Issue — 2026-08-09
Jump to: Leadership · Product · Culture · Writing · AI · Business · Data · Hiring · Career · Project Management · Industry · Security · Engineering · Testing · Tech · Operations · Decision-Making · Analytics · Metrics · Architecture · Wellness · Programming · Performance · Systems · Tools · Kafka · Remote Work · Infrastructure · Linux · Protocols · Statistics · Entertainment · Autonomous Vehicles · Mobility · Venture Capital · Privacy · Events · Sponsored · Neurotech · Venture · Management · CareerGrowth · CaseStudy · DeepDive · Git · DevEx · Communication · Wellbeing · Code Quality · Documentation · Finance · Regulation · AI Applications · Education · Retail · Legal · Policy · AI Commerce · AI Impact · AI Safety · Tech Leadership · AI Security
Leadership
The Antithesis Principle
uloza+hey@proton.me · unsubscribe
Any truth about human nature can be pointed two ways: outward as a tactic for working people, or inward as a warning about your own defaults. Smart people chase the tactic, wise people eliminate the default in themselves first: “You will not only reach the obvious conclusion that everyone who’s smart reaches, but perhaps also the non-obvious one that only the wise do.”
Why Do I Write?
uloza+hey@proton.me · unsubscribe
Writing isn’t just communication—it’s building. What breaks people is failing to build in some form, whether that’s code, docs, or a newsletter. Those who lose motivation usually stopped treating their outlet as building anything at all.
How successful companies go blind
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Successful companies can lose the ability to recognize good engineering the same way cave fish lost their sight: the environment stops rewarding the trait, so it quietly disappears. New hires learn broken systems, then replicate them through hiring panels, and careful engineers who push back get overruled until they leave or adapt.
How much should a manager code?
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Coding as a manager isn’t about hitting a time percentage. It’s about staying connected to how your systems and team actually work. The real risk isn’t coding too much or too little — it’s coding to carry delivery yourself instead of building a team that runs without you. A green contribution graph isn’t the goal. A team that barely needs you is.
How do I deal with my team member resisting a change?
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Overt resistance to change is often more useful than it looks. The person objecting out loud is still engaged, and their pushback frequently signals a flaw in the plan, an unacknowledged cost, or simply exhaustion from too many changes at once. Research shows that perceived fairness, not enthusiasm or persuasion, most reliably reduces resistance.
How do I create accountability without authority?
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Cross-functional leaders often assume that once a commitment is made in a meeting, it will hold. It rarely does without deliberate structure to support it. Accountability in a matrix organisation must be built through precise agreements, visible records, and early conversations about competing priorities. The goal is habits the whole group owns, not just the leader, so accountability survives long after any single person moves on.
The Mario meeting
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Companies plan compensation budgets more than a year in advance, long before anyone sees their review. At any given time, two budgets run at once: one being spent now, one being built for next year. Senior leaders must understand this full process, not just the end result. Most employees only see the final number, but the real decisions happen much earlier, in rooms most people never enter.
A return to two-pizza culture
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
The social physics of conversation
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Research from MIT found that how groups communicate matters more than what they say. Tracking tone, turn-taking, and who speaks to whom predicted team performance better than intelligence or talent combined. Groups where conversation flows freely between all members consistently outperform those where exchanges funnel through one central person.
How to ask for help from people who don’t know you
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Asking for help is a skill anyone can learn. The key is to think from the other person’s perspective. Make yourself worth helping by showing real proof of your work, keep your context brief, and make your request small and specific. Also make it easy for them to say no, since willing help builds lasting relationships while forced help destroys them.
The four pillars of engineering management
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Engineering managers exist because growing organizations need coordination, alignment, and direction that flat structures cannot sustain. The role varies by company and context, but the core job stays the same: make the team effective. This breaks down into four areas — leading people, guiding technical decisions, shaping product thinking, and driving delivery. Success means understanding what blocks the team and removing those obstacles.
How Meta sets up super IC teams
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Small, high-powered teams of experienced individual contributors are gaining popularity at companies trying to launch new products or transform operations. These “super IC teams” work best with four things in place: an executive sponsor with broad influence, full team dedication with no split responsibilities, a sharp focus on a real customer problem, and a handoff plan before scaling becomes urgent. Without these conditions, teams waste time on politics and lose momentum.
How to fix your all hands
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
A small problem
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
In 1982, Captain Eric Moody told passengers his 747 had “a small problem” after all four engines failed over the Indian Ocean. That understatement was deliberate: the cabin needed calm, while the cockpit got the full brutal truth. The lesson is that composure is something you broadcast, not just feel, and the message must match who receives it.
What does “playing politics” mean for software engineers?
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Playing politics at work isn’t about scheming. It means knowing who holds real power, avoiding conflict with them, and actively helping them succeed. When you disagree with a powerful person, do it privately, stay polite, and drop it when overruled. Make sure your contributions are visible, and respond quickly when senior people need help.
Effectiveness comes from developing both leadership and management skills
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Leadership and management aren’t opposites — they’re both necessary. Leadership creates direction and inspires movement, while management turns that movement into real results. One without the other leads to either vision without execution or busy work without purpose. The good news is that you’re already practicing both, since every time you influence others or make smart use of limited resources, you’re doing exactly that.
Avoid the trap of magical thinking
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Good leadership isn’t about inspiring speeches or chasing productivity hacks. It’s about understanding real business challenges, removing bottlenecks, and focusing teams on what actually matters. Magical thinking lets leaders avoid responsibility, so when things fail, they blame effort rather than direction. Strong teams built on clear goals and good communication outperform collections of star individuals.
Engineering leaders day-to-day activities
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
The best prioritization is no prioritization
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Prioritization frameworks waste time and rarely prove their worth. Instead, build faster: speed beats clever decision-making every time, and shipping more teaches you what works. Keep teams focused on one area permanently so decisions stay simple and comparable. Treat resourcing as your main lever, not constant replanning.
The antithesis principle
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Most people notice how human tendencies like needing entertainment, following charisma, or judging first impressions affect others. The smarter move is also turning that observation inward and working to eliminate those same tendencies in yourself. This is the Antithesis Principle: truths about human nature point both outward as tactics and inward as warnings. Smart people find the tactic. Wise people also find the warning.
Framework for managers starting a new role
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Starting strong in a new management role is less about moving fast and more about building the right foundation. Focus on absorbing company culture, building relationships early, and aligning with your manager on clear expectations. Resist the urge to prove yourself immediately. Instead, listen to your team, spot patterns, and let your first wins come from solving problems that actually matter.
How to think about span of control
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Span of control and team size involve real tradeoffs. Fewer management layers reduce complexity and improve clarity, but overloaded managers neglect their teams and slow everything down. Small teams of around eight people tend to move faster and work better together, while larger teams offer more resilience. As companies grow and processes mature, spans of control can increase, but mandates pushing managers to oversee 20 or more people rarely work well in practice.
No wind is a good wind
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Teams move fast but build the wrong things when leadership lacks a clear vision. There’s a simple check: ask yourself what the vision is, then ask a few engineers the same question and see if the answers match. If the vision exists but you’re not part of those conversations, push your way into the right rooms. If it doesn’t exist at all, that’s a problem worth raising loudly.
The best prioritization is no prioritization
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Before you delegate
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Delegating is hard when explaining a task feels slower than just doing it yourself. Wes shares a simple checklist of six questions that makes it easier: understand what the person already knows, share the why, gather needed resources, show what good looks like, set a clear timeline, and flag likely risks. These questions help build shared context fast and improve the quality of work handed off.
Overpromising
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Promising career rewards to keep team members happy often backfires badly. The initial boost fades fast, and if the promise falls through, trust is shattered and can take years to rebuild. Managers should only make promises they can fully guarantee, and caveat anything uncertain with honest disclaimers. Focus on what you can control, not what you hope will happen.
Engineering management after the cost of code collapsed
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Engineering management rules aren’t uniformly obsolete or valid in the AI era. The key is checking what assumption each practice rests on. If a rule depends on the cost of writing code, revisit it. If it depends on how humans coordinate, verify, or accept risk, it still holds. Code generation got cheap, but specification, judgment, and ownership did not. Whoever has to sign a decision still sets the pace.
When nothing happens
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Organizations don’t punish bad behavior — they punish bad behavior when results turn. Research shows that whether a harsh leader is seen as inspiring or abusive depends mostly on whether the team is winning. Cases like Steve Jobs, Linda Wachner, and Bobby Knight all follow the same pattern: the behavior stays constant, but the moment results drop, tolerance evaporates. The real question for any leader is how much cover strong performance is currently buying for someone on their team.
A Small Problem
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
In an organization you are always broadcasting on more than one channel at once, whether you intend to or not. Your exec team needs the unvarnished situation. The wider company needs enough to stay oriented without spinning out. Customers need a different message again. The failure mode is not the absence of a calm voice. It is using one voice for everyone: either feeding the cockpit a soothing story or dumping the cockpit’s panic straight into the cabin. Calm is contagious, and so is fear. Whichever one you transmit is the one that spreads.
How Successful Companies Go Blind
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Companies who have forgotten what it took to become successful are similar: they stop recognizing competence, because the environment has stopped expressing the trait in anyone the company hires. Call it competence blindness, which is different from incumbents who fail because they cling to the customers and margins of yesterday’s market. Firms with competence blindness do not disappear. In fact, they can survive for decades.
Mission To Metrics
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Mission to Metrics is a written cascade, running from why the org exists down to the numbers that measure whether it’s getting there. Every team’s plan is a more detailed version of one slice of the layer above. Every metric traces back to the mission. Once it’s written, anyone in the org can read the page and understand how their work connects to the company mission.
The Vital Few
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Stewardship is the actual work. It means giving your strongest people air cover, not just assignments. It means defending their focus from the steady drizzle of interruptions that drifts toward competence. It means offering them growth rather than merely more volume, and fighting the reflex to drop the next critical thing on the proven desk simply because it is proven.
Make No Assumptions
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
When that architect leaves, they are replaced by someone who aspires to operate in the same code layer, but simply cannot because they lack enough context to do so. As a result, that new replacement creates a new code layer, despite not intending to. If the team runs through a handful of folks as the new team leads struggle, it’s easy to end up with a complex code horizon very quickly.
The Art Of Dealing With Ambiguity
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Article on managing ambiguity in leadership roles Most popular from previous newsletter issue Strategies for dealing with uncertain situations
The Productivity-Experience Paradox
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
If experience and productivity really are decoupling with AI, will leaders still be able to justify caring about developer experience? I really hope so, but the case for investing in it was always that it paid off in output, and that’s exactly what’s now up for debate.
The Art Of Dealing With Ambiguity
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
The magic happens when you reframe ambiguity as an opportunity to define the problem space. Once you’ve established clear boundaries, success metrics, and constraints, the fog lifts, and the path forward materializes.
Defensible, But Not Good
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
It’s a natural human tendency to optimize for “not getting into trouble” rather than “figuring out the truth.” Even environments that pride themselves on rejecting mediocrity can fall into this trap of surface-level adequacy.
Lucas’ Laws of Project Management
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
The Conductor Developer
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
A great conductor is first and foremost a great musician. They could play the instruments themselves. That’s not why they’re standing on the podium. Their value comes from understanding the whole score. The orchestra doesn’t need the conductor because the musicians aren’t talented enough. It needs the conductor because someone has to hold the whole system in their head. Increasingly, I think that’s what great software developers are doing.
Product
Swap 30 min meetings for 3 second answers
uloza+hey@proton.me · *Sponsored* · unsubscribe
Unblocked augments source code with discussions from GitHub, Slack, Confluence, and more, so your team gets accurate answers about your codebase instantly.
This tiny, magnetic e-reader could stop you from doomscrolling
aiste.ulozaite@gmail.com · unsubscribe
The Xteink X3 is a delightfully tiny, MagSafe-compatible e-ink reader that attaches to the back of your phone like a Pop Socket.
Get Paid 30% for Sharing MyClaw
MyClaw Newsletter · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Earn 30% commission on every new customer’s first order No approval queue, minimum spend, or cost to join Share your referral link anywhere to get paid
MyClaw Platform
MyClaw Newsletter · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
MyClaw is a best-in-class AI agent platform Integrates OpenClaw, Hermes Agent, and commercial infrastructure Promises always-on agent deployment in seconds Plans to add Claude Code, Codex, and other leading agents
MyClaw AI Agent Platform
MyClaw Newsletter · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Hosts OpenClaw and Hermes agents Supports multiple AI models and frameworks Cloud-based platform for deploying AI agents
Culture
VC Catches Founder Euromaxxing
uloza+hey@proton.me · unsubscribe
My humble effort to help you start the weekend with a smile.
Writing
Don’t Be a Meat Proxy
uloza+hey@proton.me · unsubscribe
Verbose writing is often lazy writing. Copying it straight from the output you got from AI is an even worse type of laziness. Writing a response in your own words isn’t about politeness, it’s a decent certificate that you actually read.
AI
What’s Gone Wrong With AI & Labor — A Thought Experiment
uloza+hey@proton.me · unsubscribe
Coding agents haven’t replaced engineers because the software industry open-sourced every intermediate artifact, from specs to pull requests to code reviews, giving agents a training set that teaches collaboration instead of just imitation. Every profession chasing its own AI moment should be asking whether it can manufacture that same missing middle.
Escaping the LLM Coding Rat Race
uloza+hey@proton.me · unsubscribe
The AI speed race isn’t producing happier customers: 30% increase in total supply; 3% growth in utilization. This suggests the bottleneck isn’t shipping speed but effective distribution and adoption.
The AI Testing Gap Between Demo and Production
Leadership in Tech · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
A model that works in a demo and one your team trusts on release day are two different things. At the QA Leadership Summit, leaders from companies like Google Cloud and Salesforce walkthrough how they’re putting AI into testing, measuring its impact, and building reliable systems that scale. Free virtual event, July 22.
[Webinar] 8 levels of context maturity in AI-native development
Leadership in Tech · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
AI is in your engineering workflow. While the token spend shows it, the throughput doesn’t. The human is very much still in the loop, and that’s a context problem. Join live (FREE) on Jul 23 to see the 8 levels of context maturity: where most teams are stuck, what the ceiling looks like at each stage, and what it actually takes to make the most out of your agents.
Notes on the software factory
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
AI coding agents work best when given real tools like browser access, deployment pipelines, and test runners. Adding a critic agent to review work improves quality, and pairing that with more thinking time on hard problems helps further. Memory and long-running workflows are still weak points. The whole stack, from cloud infrastructure to IDEs, needs to be rebuilt around agents rather than adapted from human-centered tools.
Raise the ambition threshold
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Building software faster with AI doesn’t mean building more software. Every new system adds ongoing costs to operate, secure, and maintain, and too many obligations can crowd out the capacity to build what truly matters. AI should raise the bar for what teams attempt, not lower the bar for what gets built. Companies that use AI only to clear backlogs risk being overtaken by bolder competitors tackling harder, more valuable problems.
The new GPT-5.6 family: Luna, Terra, Sol
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
When AI Costs More Than the Engineer
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Understanding the Dynamics of the AI Ecosystem with Pace Layers
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Understand cost, adoption, and productivity of your org’s AI coding agents
Leadership in Tech · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Bills are skyrocketing but engineering leaders are flying blind. AI Coding Insights from Dash0 connects token spend to what is actually produced. Cost per merged PR, adoption across teams, and cycle time per model, with a drill-down from any number to the exact agent session behind it. Works with Claude Code and Cursor. Free until August.
The twilight of the chatbots
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
AI Bots Are Changing How Enterprises Monitor and Manage Their Infrastructure
Leadership in Tech · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
AI bots are driving more requests, more telemetry, and new security challenges across enterprise environments. But the biggest challenge isn’t detection. It’s understanding what bots are doing, which ones can be trusted, and how they affect performance, security, and customer experience. Based on a survey of 300 enterprise IT and security leaders, this report examines why organizations are investing in better visibility and real-time analytics at scale to manage AI-driven traffic.
Are we offloading too much of our thinking to AI?
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Offloading thinking to AI is increasingly common, from trivial daily choices to complex reasoning. While this can free up time and boost productivity, Yennie argues it risks eroding our autonomy and ability to think independently. Using AI to check and extend our own thinking differs from replacing that thinking entirely. The real question is whether we are automating routine work or giving away our own judgment and desires.
[Webinar] Can you prove AI is working?
Leadership in Tech · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
AI is in your engineering workflow. While the token spend shows it, the throughput doesn’t. The human is very much still in the loop, and that’s a context problem. Join live on Aug 19 (FREE) to learn how to track AI productivity, how context maturity affects where teams stall, and what it takes to get real value from your agents.
Make no assumptions
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Codebases often build up distinct layers over time as new developers add code without fully understanding what came before. AI tools are making this worse, and the same problem is now spreading to how teams reason and make decisions. AI-generated analysis can look polished while being built on flawed data, leading teams to solve problems that don’t exist. The fix is to inspect every layer of reasoning rather than accepting it at face value.
AI Companies Are Trying to Hide a Staggering Amount of Debt
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Judge approves a $1.5B Anthropic settlement over books used to train Claude
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Kimi K3, Qwen 3.8, and Anthropic’s (potential) Unravelling
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Who’s Afraid of Chinese Models?
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
AI advice made people three times less accurate but twice as confident
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
LLMs reward expertise
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Domain knowledge is the most important skill when working with LLMs. Terence Tao’s ChatGPT conversation about advanced mathematics shows this clearly. He could steer the model, spot weak answers, and suggest better directions because he already understood the subject deeply. The same applies to coding or any other field: the more you know, the more you can push the model toward what you actually want.
The AI productivity gap
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
AI speeds up coding, but coding is only a small part of most developers’ days. Senior engineers spend most of their time in meetings, reviews, and system design, areas where AI helps little. This means AI saves seniors around 15% of their time, while juniors, who code more, gain closer to 25%. Expect steady improvements, but not dramatic productivity leaps, especially from your most experienced staff.
AI’s top startups are barely publishing their research
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
What is really happening to jobs? Separating AI hype from reality
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Learn how leading teams are building and scaling AI agents in production
Leadership in Tech · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Agents can write and run code seconds before execution, often without human review. Local environments and standard containers weren’t designed for that level of autonomy. LangChain’s latest guide explains what safe agent execution requires, including isolation, permissions, observability, fast provisioning, and persistent state. Read the guide to learn when a managed sandbox makes sense, what capabilities to look for, and how to give agents the tools they need without handing them your infrastructure.
[Webinar] Measuring AI developer productivity: Can you prove AI is working?
Leadership in Tech · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
AI is in your engineering workflow. While the token spend shows it, the throughput doesn’t. The human is very much still in the loop, and that’s a context problem. Join live on Aug 19 (FREE) to learn how to track AI productivity, how context maturity affects where teams stall, and what it takes to get real value from your agents.
The lights were never off
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
The idea of the dark factory promises software development with little or no human involvement. But the teams closest to this work tell a more nuanced story. Humans still review, verify, and guide AI-generated code. The real challenge is not removing people, but replacing the work reviewers do with strong tests, simulations, automation, and safeguards.
Share your AI agent’s work with your team
Programming Digest · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Your agent generates a report, dashboard, or prototype as HTML or Markdown. display.dev turns it into a secure URL your team can open with their existing work login, comment on inline, and send back to any agent through MCP. Works with Cursor, Codex, Claude Code, and more. No viewer accounts, no seat management, no vendor lock-in.
Implementing Production Agent Memory Without Distributed Complexity
Programming Digest · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Building context and memory for production agents usually requires managing separate vector databases, caching layers, and session stores. This fragmented stack introduces latency and massive engineering overhead. Redis Iris fixes this by unifying the context layer at runtime with semantic search, persistent agent state, and caching into a single engine. Clone the demo repo to see the code in action.
How Claude code works, from tokens to agents
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Tools like Claude Code, ChatGPT, and Cursor look like magic from the outside. You type a message, and the AI reads your files, fixes bugs, runs tests. Under the hood, it’s a stack of pieces, and once you see them, the behavior of these tools gets a lot more predictable. Nemanja’s walkthrough builds that stack from scratch, starting from the simplest possible interaction and adding layers until we arrive at something like Claude Code.
Stop re-explaining your codebase to your AI agent every morning
Programming Digest · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Coding agents are stateless, your context isn’t. Jolli Memory stores agent memory as Git-backed files in your repo, so it persists across sessions and follows you between Cursor, Claude Code, Codex, and VS Code. Ships with the codebase to every teammate. Free and open source.
In-house LLM serving at Netflix
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Netflix runs its own LLM serving stack on top of vLLM and NVIDIA Triton, integrated into existing production infrastructure rather than a separate system. Key decisions around engine choice, model packaging, API design, and deployment strategies each revealed unexpected trade-offs only under real production load. A notable example is constrained decoding, where per-request CPU processing caused latency to grow linearly with batch size until a rewrite using vLLM V1’s batch-level API and C++ solved the bottleneck.
Stop re-explaining your codebase to your AI agent every morning
Programming Digest · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Coding agents are stateless, your context isn’t. Jolli Memory stores agent memory as Git-backed files in your repo, so it persists across sessions and follows you between Cursor, Claude Code, Codex, and VS Code. Ships with the codebase to every teammate. Free and open source.
Harness engineering deep dive
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
The model is just a small part of a real AI agent system. The harness — the loop, tools, sandbox, state, sensors, and guardrails — is where most of the engineering actually lives. Sensors should transform tool output into clean signals; the agent should never decide it is done on its own; and action spaces shrink through tool design, not longer instructions. The outer harness is the durable asset that compounds over time, while models get swapped and commoditized.
How LLMs figure out what you mean
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
An AI can’t work with raw text, so it converts each word into a list of numbers called a vector. Think of each vector as an arrow, where words with similar meanings point in similar directions. The model then scores how closely any two arrows align and converts those scores into percentage weights, showing how much attention each word pays to every other word.
Ship long-horizon agents that hold up in production
Programming Digest · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Long-running agents break the assumptions behind standard app infrastructure. They can run for minutes or hours, pause for human input, recover from worker crashes, and still need controls around cost, memory, code execution, and tracing. Download the engineering guide to the runtime primitives that keep production agents reliable.
Mapping out OpenAI’s startup empire
TechCrunch · via aiste.ulozaite@gmail.com
OpenAI raised a big startup fund from outside investors — $175 million in its main fund and another $114 million in special purpose vehicles. Since the OpenAI Startup Fund was launched in 2021, it’s backed more than a dozen startups (that we know of).
What is Mistral AI?
TechCrunch · via aiste.ulozaite@gmail.com
Well, it might be the only European startup that can compete with OpenAI.
Judge denies xAI’s request to block Minnesota ban on ‘nudify’ apps
aiste.ulozaite@gmail.com · unsubscribe
Despite a lawsuit from xAI, a Minnesota ban on apps that allow users to “nudify” images can move forward.
YouTuber Hank Green says his AI usage is ‘not healthy’
aiste.ulozaite@gmail.com · unsubscribe
Green offered a remarkable apology, saying that “the level of dopamine that I’ve been getting from interacting with LLMs … is not healthy for me or good for the world.”
Google nixes its Earth AI feature one day after launch, amid criticism it would spread misinformation
aiste.ulozaite@gmail.com · unsubscribe
A tool that allowed anyone to generate fake AI-generated imagery and superimpose it over real Google Earth maps quickly spurred backlash.
OpenAI reportedly finds evidence that more of its agents ran amok
aiste.ulozaite@gmail.com · unsubscribe
OpenAI has reportedly found evidence of additional agent misbehavior as it looks into the incident that occurred with Hugging Face.
Sam Altman is still making the case for parenting via ChatGPT
aiste.ulozaite@gmail.com · unsubscribe
OpenAI’s CEO seemed excited to share a “cool use case” for parents.
‘This is fine’ creator says AI startup stole his art
aiste.ulozaite@gmail.com · unsubscribe
The ad comes from Artisan, the AI startup behind billboards urging businesses to “stop hiring humans.”
Nicolas Sauvage is betting on the boring parts of AI
aiste.ulozaite@gmail.com · unsubscribe
The portfolio he has assembled since 2019 is dotted with technologies that have become more widely interesting to VCs over the last year:
The AI safety test is becoming a safety risk
aiste.ulozaite@gmail.com · unsubscribe
AI agents are escaping cybersecurity testing environments and reaching real-world systems, raising questions about whether safety infrastructure, industry standards and regulation can keep pace with increasingly powerful models.
OpenAI acquires presentation startup NextSlide
aiste.ulozaite@gmail.com · unsubscribe
NextSlide says its team members are now working on ChatGPT.
Cloudflare launches Kitesurf, a browser built for AI agents
aiste.ulozaite@gmail.com · unsubscribe
Kitesurf is a cloud-hosted browser designed for AI agents instead of people. It uses less computing power than Chromium for common automation tasks, helping developers build browser-based AI agents more efficiently.
Free Gauntlet AI Night School: MCP Factory — Build Any MCP You Want
Pointer · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Aaron Gallant breaks down MCP Factory, a repository that lets you spin up any MCP server you want instead of hand-building each integration from scratch. MCP servers are how AI agents plug into real tools and data, and Aaron walks through how the repo is structured, how to go from an idea to a working MCP fast, and how generating them changes what you can wire your agents into. With the next cohort approaching, it’s also a clear look at the kind of work Gauntlet engineers do every day. You’ll learn: (1) Why MCP servers matter for connecting agents to real tools and data. (2) How MCP Factory is structured as a reusable repository. (3) How to go from an idea to a working MCP quickly. (4) Practical patterns you can take back to your own projects.
Reverse Engineering ChatGPT Web: How OpenAI Built for a Billion Users
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
On the surface it seems like a simple interface, but after digging in it’s anything but. I spent days reverse engineering their web app by digging through the page source, bundled code, and network requests to understand how it was built.
Why Stitched-Together AI Architectures Backfire
Pointer · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Building context and memory for production AI agents usually means stitching together a fragile tool zoo of vector dbs, sync pipelines, and external session caches. This fragmented architecture breeds massive engineering overhead. Redis Iris fixes this by unifying context orchestration, persistent agent memory, and semantic caching into a single, low-latency platform. Try the hands-on tutorial.
Using Local Coding Agents
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
This article is a tutorial on setting up a production-ready coding agent with a fully local stack. We will use a locally served LLM together with a local coding harness that can read files, make edits, run commands, and verify changes as shown in the figure above.
[Webinar] 8 Levels Of Context Maturity In AI-Native Engineering
Pointer · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
AI is in your engineering workflow. While the token spend shows it, the throughput doesn’t. The human is very much still in the loop, and that’s a context problem. This free webinar maps the 8 levels of context maturity: where most teams are stuck, what the ceiling looks like at each stage, and what it actually takes to make the most out of your agents. Join live July 23 (FREE).
The State Of AI Impact In Engineering: Q2 2026
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Engineering leaders are now under immense pressure to justify exponentially-increasing AI budgets. The data from our Q2 report reveals that while AI is delivering objective gains in velocity, those gains are highly uneven.
The New Rules Of Context Engineering For Claude 5 Models
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Here’s what we’ve learned about prompting this new class of models, and how you can utilize it to update your context engineering. We’ve put these best practices in claude doctor, use the command /doctor in Claude Code to rightsize your skills, and CLAUDE.md files.
How Building Software Is Changing At Anthropic
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
A deepdive on what’s changed in how the leading AI lab makes software. Ever more code review and testing is done by AI, two-pizza teams very much alive, and more.
The New Rules Of Context Engineering For Claude 5 Models
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Context engineering practices for Claude 5 models Updated rules and best practices for prompt engineering Optimization techniques for AI model interactions
[Live Session] Can You Prove AI Is Working?
Pointer · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
AI is in your engineering workflow. While the token spend shows it, the throughput doesn’t. The human is very much still in the loop, and that’s a context problem. Join live on Aug 19 (FREE) to learn how to track AI productivity, how context maturity affects where teams stall, and what it takes to get real value from your agents.
My Agentic Coding Setup, July 2026
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Things are changing very fast, and I have no illusions that this will be a timeless nugget of wisdom. But talking to some friends and scrolling through X posts, it does seem that I might have some techniques in play that are not widespread.
Gauntlet AI Night School — On Demand: Watch Every Session, Free
Pointer · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Night School is Gauntlet’s free education series on what AI-native work actually looks like — no hype, no vendor pitches, just engineers and operators showing how they really build. The full library is now available on demand, so you can watch any session whenever you want. Go deep on planning, specs, and decision logs that separate AI engineers from vibe coders. See how a team ships go-to-market work with AI in the loop, how to build any MCP you want, and how to make a real game with AI-generated art and no dev background. Every session is a practical, over-the-shoulder look at how AI changes the work — recorded and ready to watch on your schedule.
Building An Advanced Agentic Harness
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
In this post we upgrade every piece of our basic harness toward that production shape, without hiding any of the mechanics behind a framework. The guiding question for the whole exercise is a simple one: How do you turn a single LLM call into a reliable system that can plan, act, recover, and prove it did the right thing?
My Agentic Coding Setup, July 2026
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Practical agentic coding setup from July 2026 Real-world implementation and tooling choices Current best practices for AI-native development
Goldman Sees AI Reshaping Indian Work
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Goldman Sachs estimates generative AI could replace 8–12% of India’s non-agricultural jobs while complementing 42–48% by automating routine tasks and boosting productivity. Services such as education, media, finance, and professional work face the most exposure, while construction and other physical occupations remain relatively insulated, pointing to broad labor reallocation rather than mass unemployment over time.
AI Traffic Favors Buying Pages
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
WebFX analyzed nearly 600,000 AI referral sessions across 2,500 URLs and found that AI sends users mostly to decision-stage content, not awareness pages. Homepages led with 31.3%, while product, service, blog, and FAQ pages captured most remaining traffic. Pages with specific facts, pricing, citations, and expert signals performed best, suggesting AI rewards content that helps buyers act.
AI Cracks Outsourcing’s Revenue Engine
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Indian IT outsourcing giant Infosys cut its fiscal 2027 growth forecast to 1.5–3.0%, acknowledging that customers are shifting technology work toward AI. The warning matters beyond one company: outsourcing was built on expanding teams and billable effort, but AI lets clients internalize more capability, compress project scope, and spend less on labor-heavy services, weakening the industry’s traditional revenue engine.
Agents Scale Sleep Discovery
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Researchers built AI Sleep Co-Scientist, an expert-guided system that analyzed roughly 124,000 sleep studies and more than 50 TB of physiological signals. By automating labor-intensive preprocessing, hypothesis development, and statistical analysis while keeping humans in review.
Amazon Tests Health Agents
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Amazon introduced PatientAgentBench, a clinician-vetted framework that creates synthetic patient records and multiturn conversations to test healthcare AI agents on safety, triage, workflows, and task completion. Across thousands of evaluations, even frontier models missed crisis resources, fabricated clinical details, and mishandled routine requests hiding serious risks.
Zuckerberg Bets on Personal Agents
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Meta CEO Mark Zuckerberg predicts billions of people will use personal AI agents within five years to manage finances, health, relationships, and households through platforms like WhatsApp. He says these agents will underpin Meta’s next products and revenue, even as AI infrastructure spending surges, free cash flow falls 91%, and Reality Labs’ cumulative losses reach roughly $88 billion.
The Box-Checking Agent
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Mark Cuban says his Shark Tank company Rebel Cheese is saving $50,000 a month with a simple AI agent that photographs shipping boxes and invoices, checks sizes against carrier price lists, and automatically files credit requests when charges are wrong. The lesson is not flashy automation, but agents quietly finding money in boring operational gaps.
Hiring’s Algorithmic Blackball
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Stanford-led researchers found AI hiring vendors can turn one bad assessment into repeated rejection, as stored scores follow applicants across employers using the same platform for up to 330 days. Their Pymetrics audit found lost job advances and racial disparities, warning that opaque hiring systems can quietly blackball candidates before any human review, even across unrelated roles.
Agentic Commerce Takes Cart
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
McKinsey says AI agents could mediate $3 trillion to $5 trillion in consumer commerce by 2030, moving shopping from human persuasion to machine-readable competition. As agents compare prices, fill carts, and eventually negotiate with store bots, retailers that expose inventory, pricing, and return data through APIs will win while human-only marketing loses power inside retail.
AI’s Missing Consumer Problem
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Consumer intelligence company CivicScience CEO and founder John Dick says in Cannes that marketers are celebrating AI while ignoring consumers who remain skeptical, scared, and worried about jobs, energy costs, and the economy. His argument reframes AI as a trust problem, making first-party data more valuable and pushing brands from performance obsession toward a more balanced brand-building strategy in media planning.
Home Depot Replaces Menus
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Home improvement giant Home Depot is replacing rigid phone menus with voice agent that lets callers simply explain what they need. In a 50-store pilot, the system handled intent in under 10 seconds and proved four times faster, while also building carts, starting service requests, resolving common questions, and routing customers to associates when needed.
Agents Turns Goals Into Experiments
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
PNNL researchers built an AI agent system that lets scientists describe experiment goals and turns them into robot-ready instructions for Big Kahuna. By coordinating specialized sub-agents, it replaces weeks of scientist-engineer translation with automated workflow design, helping labs run five to 10 times more chemistry experiments while humans guide strategy and robots handle execution.
Agents Need Trust Infrastructure
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Linux Foundation is launching ANS, a DNS-style identity layer for AI agents, so enterprises can verify who an agent represents, what permissions it has, and whether its code and history remain authentic. The move tackles a real control-plane gap as agents cross tools and companies, though DNS security limits and competing standards still leave adoption unsettled.
Alibaba Reopens the AI Race
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Alibaba released Qwen3.8-Max, its largest and most capable model yet, claiming performance near Anthropic’s Fable 5, OpenAI, and China’s Kimi K3. Arena rankings place it just behind top Claude models, while Alibaba plans to release open weights next week. The launch deepens US-China competition and underscores China’s push to make advanced open-weight AI widely available.
AI Runs the Dental Office
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
In an a16z interview, Lassie’s founders explain how the company turned a dentist’s 200 monthly hours of paperwork into an AI agent that automates about 98% of billing and administration. He said AI software will not merely organize business data, but perform labor directly and help understaffed small businesses run themselves at scale.
Retail AI Looks Away
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
A Columbia Law School study says Amazon’s Alexa for Shopping and Walmart’s Sparky can identify suspicious “Made in USA” claims, yet neither retailer systematically flags them. Researchers argue the gap reflects business incentives: unless false labels create financial, regulatory, or reputational costs, detection alone changes little, leaving regulators to ensure AI serves shoppers rather than sales.
Hermes Cleans Its Own Skills
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
In Peter Yang’s interview, Nous Research co-founder Karan Malhotra explains that Hermes not only creates its own skills, but maintains them through Hermes Curator, a scheduled background process that reviews memories and skills for bloat, redundancy, and inefficiency. Because the system is open source, users can define “slop” themselves and rewrite the cleanup loop around their own standards.
METR Calls for Outside Investigations
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
METR, one of the independent evaluators of frontier AI risk, says serious agent failures should be investigated by outside experts, not only the labs involved. After documenting 44 incidents, including deception, sandbox escapes, and the Hugging Face breach, it argues third-party access to models, logs, staff, and training data is needed to uncover real causes and verify that fixes work.
The AI Ecosystem Expands
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Goldman Sachs says AI infrastructure investment is spreading beyond hyperscalers as enterprises, governments, and neocloud providers expand computing capacity. July projects included Meta and BlackRock’s Texas data center, sovereign initiatives in Saudi Arabia, Croatia, and Indonesia, and new enterprise deployments at Starbucks and Revolut. The report says AI is becoming a broader ecosystem, though its survey covers only selected announcements.
Fortnite Teaches Agents the Lobby
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Epic Games appears to have added a “Default Agentic Control” file that explains Fortnite’s lobby interface to an automated testing tool or AI agent. The file could help internal bots navigate menus, understand UI behavior, and identify regressions during playtesting, though its exact purpose remains unconfirmed.
AMD Sees Agents as Workers
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
In a Forbes interview, AMD executive argues AI agents are shifting enterprise software from reactive assistants to autonomous workers, promising far larger productivity gains but creating cost, security, and infrastructure demands. He says companies should adopt hybrid architectures, running routine workloads locally on agentic PCs while reserving cloud frontier models for complex tasks, and start with targeted, governed pilots.
World Bank Calls AI a Lifeline
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
World Bank says developing economies could compress a century of progress into a decade by adapting small, low-cost AI tools for healthcare, education, justice, and farming. Only 4.5% of jobs are exposed, versus 14.2% in rich countries, but governments must close electricity, connectivity, device, and skills gaps while guarding against inequality, misinformation, and political repression.
Agent Plugins Go Cross-Platform
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Vercel and partners including AWS, GitHub, Microsoft, OpenAI, and Cursor launched Agent Plugins 1.0.0, an open standard for packaging Agent Skills and MCP servers into one portable format. Using a simple plugin.json manifest and fixed folders, authors can build once for compatible clients, including ChatGPT, Codex, Cursor, GitHub Copilot, Kiro, and VS Code, without repackaging components separately.
Motor Design Drops to 20 Minutes
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Atlas Motion emerged from stealth with $11.5M in funding and says its AI can compress motor design from roughly two months to about 20 minutes, then move new designs into scaled production within six weeks. Founded by Tesla and defense-tech veterans, the startup is already shipping 10,000 motors monthly and targets 40,000 by December, mainly serving defense customers.
Cloudflare Tracks AI Brand Visibility
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Cloudflare is launching its AEO Visibility Dashboard to show brands whether AI assistants such as ChatGPT and Gemini cite, mention, or recommend them. Unlike its Agent Readiness tool, which checks crawler access, the new dashboard measures what happens after content is read.
AI Is Taking Coworkers’ Tasks
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Nearly half of employed U.S. adults now use AI at work, according to an Epoch AI-Ipsos poll of roughly 1,100 workers. One in five said they use AI for tasks previously given to colleagues or contractors, while document reading and data analysis were common uses. Researchers say AI is reshaping individual tasks faster than replacing entire jobs.
Reddit Moves Against AI Scraping
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Reddit is tightening Old Reddit as it tries to stop AI companies and automated systems from scraping its valuable user-generated data. The legacy site will require logins, while bots and moderator workflows move to newer infrastructure. Reddit hasn’t revealed all planned restrictions yet, but the changes show how aggressively platforms are starting to protect content from AI crawlers.
Grokipedia Stops Updating
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Elon Musk’s Grokipedia appears to have largely stopped processing edits since late April, according to a Lawfare analysis of 34,519 pages and 225,496 suggested changes. No corrections were accepted or rejected over the past three months, while user activity dwindled and its live edit feed broke earlier. Musk has not publicly mentioned the Wikipedia rival since February.
Business
ARR Doesn’t Mean What It Used To
uloza+hey@proton.me · unsubscribe
ARR is now the most distrusted metric in the room because different definitions exist: is it revenue or GMV? Is it recurring? Is it quarterly × 4? Or 365 × daily revenue? Founders who spell out their ARR math up front build more credibility than any growth rate can.
Zombie companies
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Large companies can keep growing, renewing contracts, and reporting strong results long after their ability to build a relevant future has gone. Past advantages like brand recognition, contracts, and switching costs keep money flowing, but they say nothing about whether the business could earn its position again today. Meanwhile, challengers need less capital than ever to pick apart the most valuable pieces, and AI is making it cheaper for customers to inspect, compare, and leave.
How successful companies go blind
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Reflecting on my failure to build a billion-dollar company
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Gumroad started as a weekend project that quickly attracted millions in venture funding, but growth stalled and led to laying off 75% of the team. Rather than shut down, the founder chose to keep the business running for the creators who depended on it. Over time, the company became lean and profitable. Sahil now measures success by the value created for others, not by chasing a billion-dollar valuation.
Unlock access to startups and investors at TechCrunch All Stage
TechCrunch · *Sponsored* · via aiste.ulozaite@gmail.com
TechCrunch All Stage sponsorships offer unparalleled access to tech’s top innovators. Through speaking opportunities, interactive roundtables, and customizable activations, position your brand at the forefront of the industry. Gain valuable leads, media exposure and lasting ROI.
We’ll take it: a TikToker rallies pledges to buy Spirit Airlines after its abrupt weekend collapse
aiste.ulozaite@gmail.com · unsubscribe
Within hours he’d thrown up a website — a janky, one-hour job, by his own admission — and by Sunday, 36,000 “founding patrons” had pledged nearly $23 million, crashing his servers in the process.
Amazon opens up its global logistics network to all businesses
aiste.ulozaite@gmail.com · unsubscribe
The new service, called Amazon Supply Chain Services, pits the e-commerce giant directly against UPS and FedEx.
Booking Wants the Whole Transaction
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Booking Holdings says it does not want to remain merely an AI travel recommender. It wants to control the entire transaction chain, from inventory and identity checks to payments, supplier settlement, cancellations, refunds, and support. AI helps plan trips and lower operating costs, but Booking’s strategy is to own the infrastructure that actually completes and monetizes travel.
Data
Data Lessons From Inside Meta
uloza+hey@proton.me · unsubscribe
Interview with insights about working at Meta. Key takeaway on data quality: early schema enforcement and strong typing upstream made column-level lineage and privacy work possible. Teams are growing larger as middle layers are removed and managers become individual contributors—raising questions about scaling management without weekly 1:1s.
Hiring
Real-time feedback in every interview
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Ending interviews with direct, honest feedback turns a one-way process into a real conversation. Candidates rarely get it, so they almost always appreciate it. Giving specific feedback on candidates’ capabilities also allows them to correct genuine misunderstandings, leading to better hiring decisions. Mark suggests focusing on what someone said or did, not their personality traits.
Career
The laws of this world (game theory)
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Hard work alone rarely leads to success because effort without the right position, visibility, and problem selection produces little return. The market only rewards outcomes, not hours spent. Most people follow an outdated model that no longer reflects how success actually works. Fixing your direction, making your competence visible, and solving problems the market genuinely needs will matter far more than working harder.
A practical guide to fast onboarding
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Sort incoming knowledge into three buckets: facts, processes, and concepts Concepts are most powerful because they transfer across roles and domains Structured categorization enables faster contribution in new roles
Advice to new grads: Take any job you can get in the industry
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Getting into the software industry matters more than getting the perfect role right away. Taking jobs in support, QA, or technical writing puts you close to real problems and real people. Relationships built through good work open doors that portfolios alone cannot. Proximity to the right people and problems is what moves careers forward.
Growing as an engineer in a world of AI
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Norberto argues that leaning on AI to generate code without understanding it stunts your growth as an engineer. The learning that used to happen through struggle, repetition, and debugging is exactly what AI removes. To grow, he suggests treating AI as a thinking partner rather than an answer machine: ask why, type code out instead of pasting it, read the foundational books, and stay close to the hard conversations AI cannot join.
What Does “Playing Politics” Mean For Software Engineers?
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Playing politics isn’t about plotting and scheming, and it isn’t just about being a friendly, likeable person (although that helps). It’s about figuring out how your company actually operates: who makes the decisions, who gets consulted, what behavior gets rewarded, and so on. The most basic way to do that is to figure out who is powerful, get out of their way, and (if you can) help them get what they want.
You Don’t Have To Be Smart If You Can Think Clearly
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
The difference between a “smart” engineer and a “strong” engineer is how they react to problems that aren’t solved instantly. A smart engineer might flail and struggle, hoping to find that flash of insight that eluded them; a strong engineer will have some process for methodically plodding away.
How I Find Problems To Solve As A Staff Engineer
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
How do you find problems worth working on? a senior engineer I mentor asked me recently. He’s trying to make the jump to staff engineer and realized that the role isn’t just about doing the work he’s assigned. He also needs to get involved in figuring out what his team and org should be building.
How I Find Problems To Solve As A Staff Engineer
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Staff engineer role requires identifying high-impact problems Problem discovery strategies for senior technical roles Finding meaningful work at the staff engineer level
Project Management
Laws of project management
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Lucas presents ten practical laws for managing projects successfully. He explains that projects succeed by solving real problems, adapting to new information, and delivering working solutions early. The laws emphasize flexible planning, clear priorities, teamwork, operational readiness, and testing important assumptions first. Together, they provide simple guidance for building products that meet user needs while balancing business goals, time, scope, and quality.
Industry
Zuckerberg ‘Admits’ Meta’s Layoffs Were Ineffective
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
South Korea to spend $1T on more memory chip production
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Memory Prices
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Stripe, Advent offer to buy PayPal for more than $53 billion
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Nvidia, CoreWeave, and Nebius: Inside the Circular Financing of the GPU Boom
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Irish datacenters now guzzle 23% of the country’s electricity
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
The Future Worth Building Is Human
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Banning AI in Law School: We’ve Seen This Before
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Security
Espionage Against the European Parliament
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
How We Tricked GitHub’s AI Agent into Leaking Private Repos
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Staying one step ahead of cyber attacks
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Soatok’s Informal Guide to Threat Models
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Lessons Learned from CISA’s Recent GitHub Leak
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
The Cheapest Way Into Your Business Isn’t Malware. It’s a Phone Call.
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
OpenAI’s accidental cyberattack against Hugging Face is science fiction that happened
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
What a Law Firm’s Ransomware Nightmare Can Teach Your Startup
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Hacker wipes Romania’s entire land registry database
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Read This Before You Buy That TV Streaming Stick
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
My security camera shipped a GitHub admin token in its login page
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
I inspected my take-home interview project. It was a whole operation.
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
A fake recruiter on LinkedIn sent Appaji a Python developer job offer with a take-home coding test hidden inside a zip file. The project looked clean at first, but running a simple directory listing revealed pre-loaded Git hooks designed to silently download and execute malicious scripts the moment any Git command ran. The scripts then installed Node.js in the background and ran obfuscated code targeting crypto wallets, with unique tracking IDs assigned to each victim.
This ‘adversarial’ pattern can prevent surveillance cameras from detecting you
aiste.ulozaite@gmail.com · unsubscribe
A security researcher has designed an algorithm that can create computer-generated patterns capable of hiding people, faces, and vehicles from detection by surveillance cameras.
Google’s top hacker hunter explains why hacking groups get codenames
aiste.ulozaite@gmail.com · unsubscribe
Google recently changed how it refers and assigns names to hacking groups. TechCrunch spoke with one of the world’s foremost experts on tracking hackers to understand why companies give hackers codenames.
Computer maker Framework notifies ‘all customers’ of a data breach
aiste.ulozaite@gmail.com · unsubscribe
Framework told “all” of its customers that hackers accessed their names, email addresses, phone numbers, and physical addresses in a data breach.
Security researchers scanned the Polish web and found courts, hospitals, and airports at risk of hacks
aiste.ulozaite@gmail.com · unsubscribe
Researchers found common points of failure, like software used to organize and display web content, could have allowed hackers to run riot through government websites.
Benford’s Law: The Strange Law That Catches Financial Fraudsters
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
In almost any large dataset drawn from the real world — populations of cities, lengths of rivers, prices, earthquake depths, company revenues — the leading digit is not uniformly distributed. The digit 1 appears as the first digit about 30% of the time. Not 11%. Nearly three times the naive expectation. The digit 9 appears less than 5% of the time.
Black Hat Builds Immune System
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
At Black Hat USA 2026, the Open Secure AI Alliance expanded its enterprise security push with SAFE guidelines for sharing AI incidents and near misses, plus open tools such as Nvidia OpenShell for sandboxing agents. The initiative aims to give companies inspectable, self-hostable defenses spanning identity, model safety, scanning, and agent harnesses, while reducing token costs and vendor dependence.
Engineering
How to write an effective software design document
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Writing a design doc before coding saves time by forcing clear thinking and helping teams coordinate. It should focus on high-stakes decisions where getting things wrong is costly, not minor choices that are easy to fix later. Key sections include goals, background, diagrams, security, and open issues. The right length depends on project complexity, team size, and risk.
Perfection is not over-engineering
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Over-engineering means solving the wrong problem, not caring too much about quality. A perfect solution exists when requirements are clear enough that only one answer fits. The real culprit behind bloated systems is poor requirements gathering, not ambition. Get every constraint on the table and the right solution becomes obvious.
How I find problems to solve as a staff engineer
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Finding good problems to work on comes from paying close attention to what people struggle with day to day, not from scheduled thinking sessions. Letting issues accumulate over time reveals patterns: separate complaints often share a common root cause. Testing ideas through prototypes and conversations helps confirm whether a solution is real or just elegant on paper. Over time, solving the right problems builds trust and opens doors to shaping what a team builds next.
How to measure engineering
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
ValueSum metric tracks value delivered over time by scoring shipped items Works across product, engineering, and design teams Enables learning by rescoring features months later to validate predictions
Beyond happy path engineering: Time
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Gabor explains why time is more complicated in production systems than it first appears. He distinguishes wall-clock, monotonic, logical, and business time, showing when each should be used. He also covers clock drift, expiration, scheduled jobs, time zones, competing updates, and reconciliation. The main lesson is to treat time as a design boundary and make its assumptions explicit.
Testing
Still writing tests manually? Meticulous AI is here.
Leadership in Tech · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Most teams are forced to make the tradeoff between better coverage and more maintenance effort. But top engineering teams like Notion, Dropbox, Wealthsimple and LaunchDarkly have discovered a new testing paradigm. Built by ex-Palantir engineers, Meticulous creates and maintains E2E UI tests that covers every edge case of your web app without any developer effort - making it the only tool to improve both product quality and dev velocity.
Prefactoring: Clear The Way For Your New Feature
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Prefactoring (short for “preparatory refactoring”) is the practice of reworking existing code to make it more suitable for an upcoming change before you actually implement the new functionality. Instead of cleaning up code as an afterthought or trying to force a new feature into an incompatible structure, you restructure the codebase first.
How Miro Builds And Tests Agentic Features At Scale
Pointer · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Miro’s AI team was building a new AI architecture across nearly 250 microservices, and one shared environment meant tests constantly collided. With Signadot, they moved to verifying changes in parallel with lightweight environments in the cluster. Agents drive full end-to-end tests against live services and iterate before a PR opens. Engineers get verified working code without waiting for slots.
Tech
Nano Banana 2 Lite
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
The economic benefit of refactoring
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
An AI agent built a 150,000-line application, but its data access layer ballooned into a single 17,000-line file with massive repetition. To test whether refactoring helps AI agents work more efficiently, the same code change was measured before and after each refactoring step. After 15 steps, the tokens needed to make that change dropped by 83%, because the agent could read fewer, better-organized files instead of scanning one giant one.
London Gatwick introduces robotic parking service
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Google Discloses $94.1 Billion in SpaceX Stock
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Any sufficiently advanced technology is indistinguishable from a religion
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Two-way doors sometimes lock behind you
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Reversible technical decisions become costly to undo as systems accumulate dependencies Optionality requires active maintenance and intentional design Suggests documenting assumptions, estimating migration costs, and testing exit plans
How big is a Git commit?
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Git stores data as compressed objects, so commit sizes vary based on how compressible your files are. Small commits add a few kilobytes of overhead, while larger text files can compress down to around 10% of their original size. Git also packs objects together over time to save even more space, making it very storage efficient overall.
Meet alice. Alice is impatient.
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Users and operators often measure service performance very differently. Operators count each outage or slow request once, while users experience long outages repeatedly, making them feel much worse. This is called the inspection paradox: users see a time-weighted version of your metrics, so a one-minute average recovery time can feel like an hour to them. Long tail events matter far more to users than standard averages suggest.
A peek into reddit’s anti-spam internals
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
A bug in Reddit’s spam filter briefly exposed internal removal reasons to a moderator in 2021. The leaked messages revealed several tools Reddit uses behind the scenes, including Google’s Perspective API for spam scoring, a rules engine called Spamurai, domain bans, and even URL inspection that scans linked pages for known spam patterns.
Who needs an “Uber for bodyguards”? Protector offers on-demand security details
TechCrunch · via aiste.ulozaite@gmail.com
Protector is a startup offering users the opportunity to hire their very own security detail. The app is getting plenty of attention on social media, much of it negative, which may be the point — after all, the startup’s adviser, Nikita Bier, has written about the appeal of rage bait app ideas.
How to rid yourself of the new Apple Mail design
TechCrunch · via aiste.ulozaite@gmail.com
Amanda discussed the new design of Apple Mail in iOS 18 — specifically, how to return your version of the app to the old design, and how to remove some additional visual clutter.
Google Using Pixel Phones to Monitor NYC Subway Track Sounds
TechCrunch · via aiste.ulozaite@gmail.com
Google has been working with New York City’s Metropolitan Transportation Authority to experiment with using consumer tech as a supplement to subway track inspections. So far, they’ve secured Pixel smartphones to subway cars on the A train, where they listen for suspicious sounds.
Ouster’s new color lidar is coming to replace cameras
aiste.ulozaite@gmail.com · unsubscribe
A sensor that can simultaneously capture depth and image data has long been a “holy grail,” Ouster CEO Angus Pacala told TechCrunch.
Historian Jill Lepore says Silicon Valley misreads science fiction and undermines democracy
aiste.ulozaite@gmail.com · unsubscribe
On the latest episode of Equity, we spoke to Jill Lepore about “government by machines” and why Elon Musk is a bad science fiction reader.
Planned Amazon data center could become the biggest climate polluter in the U.S.
aiste.ulozaite@gmail.com · unsubscribe
As part of a planned Texas data center, Amazon is investing in an on-site power plant that could reportedly become the largest source of climate pollution in the United States.
The Economic Benefit Of Refactoring
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
The goal of refactoring an agentic code base is to spend tokens now in refactoring to make token consumption for future work lower. An experiment should be able to show that as this file was refactored the token cost of making separate feature implementations in this code base would decrease.
How’s Linear So Fast? A Technical Breakdown
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
A few milliseconds is all it takes to update an issue in Linear. A traditional CRUD app doing the same thing takes about 300ms. How do they do it? There’s no secret silver bullet to performance. The reality is that it’s built from the ground up on the right foundation, then improved by countless decisions. My goal is to walk through some of the techniques that make Linear feel the way it does and help you implement the same.
Microsoft Softens Its AI Bill
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
To soften the reported cost of its AI buildout, Microsoft is extending the estimated life of data centers and offices from 15 to 25 years, shifting more leases outside reported capex. The accounting change does not reduce actual investment: 2026 spending remains about $175 billion, while fiscal 2027 capex is still expected to rise further.
Operations
Code yellow, code red
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
James shares his experience of how repeated outages were eroding customer trust, leading to a formal company-wide Code Yellow, targeting eight consecutive weeks of zero downtime. The effort involved auditing the entire stack, fixing rate limiting, adding circuit breakers, resolving memory leaks, and overhauling incident processes. The target was hit. Code Yellows work best with a clear problem statement, measurable exit criteria, full organisational transparency, and genuine authority to reprioritise work across teams.
Notes on incidents
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Most incidents resolve without any action, and rushing to fix things often makes them worse. The best first step is to do nothing, take a breath, and observe. Simple fixes like disabling a feature flag usually solve the problem, but only if you know the system well enough to act decisively. Fixing incidents earns goodwill, but it is not a reliable path to lasting influence.
Decision-Making
The 37% rule
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
There is a mathematical strategy to help make better decisions when choosing from a sequence of options you cannot revisit. It works by spending the first 37% of options on observation only, then picking the next option that beats everything seen so far. This gives you roughly a 1 in 3 chance of finding the best option. The key lesson is simple: calibrate long enough to set a benchmark, but not so long that you miss your best choice.
Analytics
How much of your traffic is actually AI bots? Most teams are guessing low
Leadership in Tech · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Industry benchmarks put automated traffic at over half of all web activity, yet 82% of enterprises estimate AI bots account for under 25% of theirs. That blind spot skews your analytics, degrades cache efficiency, and buries real signals in telemetry noise. Based on a survey of 300 enterprise leaders, this report breaks down how engineering teams are getting real-time visibility into bot traffic at scale, without blocking the automation they depend on.
Metrics
Please don’t smooth the metrics
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Smoothing metrics helps investors spot long-term trends, but operators should avoid it. It hides the very spikes and dips that signal real problems. A worked example shows how smoothed churn masked a sales team expanding to the wrong customers, delaying a critical board conversation by six months while the damage grew. Raw, volatile numbers tell the truth.
Architecture
A simple view on progressive software architecture
Leadership in Tech · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
The way software is built is changing. Yet, architecture thinking is more important than ever. Learn about key principles and frameworks we believe are critical to thriving in this new, chaotic world.
Building Ask DoorDash: A Platform For Building and Evolving Agents
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
We built Ask DoorDash on a common platform that lets domain teams build and evolve their agents without rebuilding the systems beneath them. We judged the platform by two practical outcomes: how quickly teams could add features and domains, and how quickly they could evaluate and release improvements to cost, quality, and latency.
The Bedrock of Software Design
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Nothing has influenced the way I design software more than algebraic data types. It sounds overly academic—the kind of term that makes people roll their eyes or scares them away altogether. But honestly, it only sounds scary. The idea itself is quite simple, and learning it showed me a fundamentally better way to build software.
Wellness
Finding calm in life and work
Leadership in Tech · via aiste.ulozaite@gmail.com · unsubscribe
Anxiety measurably reduces performance, pushing people to bottom third of peers Modern work culture rewards busyness over output, triggering chronic low-grade stress Simple habits like clear boundaries, movement, social connection, and meaningful work enable calm
Programming
Small programming tricks
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Small tricks and shortcuts add up to meaningful gains in engineering productivity. Things like fuzzy shell history search, git pickaxe, or knowing a single command to spin up a local server can save real time without requiring deep background knowledge. The same applies inside companies, where knowing who to ask or where to find a doc can be just as valuable. Sharing one tip a day with your team is a simple way to spread this kind of knowledge.
Running an obfuscated bash script from Uniqlo’s tee
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
A Uniqlo t-shirt made in partnership with Akamai contains a real bash script printed on the back, encoded in base64. After carefully transcribing and decoding it, the script turns out to be a harmless Easter egg that displays a colorful animated “Peace for All” message in the terminal. Akamai designed the shirt as a nod to the early internet era, with the code serving as a reference to Linux.
The depth-first search pattern: Exploring trees and graphs
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Depth-first search is a recursion pattern for exploring trees, graphs, and matrices. It works by going as deep as possible along one path before backtracking and trying another. To avoid infinite loops in graphs, you track visited nodes and skip them on future visits. Depth-first search fits problems where you need to find paths, check reachability, or explore all possibilities through a structure.
Everyone should know SIMD
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
SIMD lets a CPU process multiple values at once instead of one at a time, turning slow byte-by-byte loops into chunk-based operations that run 4x to 16x faster. Despite its reputation for complexity, common SIMD code follows the same five steps every time: broadcast constants, loop over chunks, operate on all values at once, reduce the result, then handle leftovers with a normal loop. Once you learn that pattern, writing SIMD is nearly as simple as writing a regular loop.
Performance
When ‘if’ slows you down, avoid it
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Branch mispredictions slow CPUs down because they force the processor to discard queued work and start over. Writing branchless code removes unpredictable jumps and lets the CPU run at full speed. A simple filter loop rewritten to use a conditional increment instead of an if statement ran nearly 10 times faster on an Apple M1. Compilers cannot make this change automatically because they cannot safely assume unconditional memory writes are always allowed.
Quadrupling Code Performance With A “Useless” If
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
We don’t expect too many chunks, so next_j[i][j] is quite likely to just be equal to j. If we could tell the CPU to predict that j stays intact, the loop would become throughput-bound rather than latency-bound.
Everyone Should Know SIMD
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
SIMD has a reputation for being complex. I’ve met many very good software engineers who dismiss it as something too complex to learn or a niche optimization meant for only the highest-performance software, not useful in everyday programming.
Every Byte Matters
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Farid argues that classic analysis isn’t enough. The size of your data matters as much as your algorithm. CPUs pull memory in 64-byte “cache lines,” so how tightly your data packs into those lines impacts speed.
Systems
Virtual memory from first principles
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Abhinav provides a practical introduction to virtual memory, explaining its core mechanisms, including address translation, demand paging, copy-on-write, memory mapping, and page reclaim. He also explores their impact on system performance, memory management, and observability, equipping readers with a mental model for building and debugging data-intensive systems.
Tools
Good tools are invisible
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Good tools should be invisible. When a tool starts creating friction, that’s a flaw, not a feature. Yet many people reframe their tools’ weaknesses as fun puzzles to solve, then mistake that feeling of cleverness for real productivity. The best sign a tool is working for you is that you stop noticing it entirely.
Review The Actual Change, Not The File List
Pointer · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
AI writes more code than ever. Reviewing it shouldn’t mean scrolling forty files in alphabetical order. CodeRabbit Review reorganizes any pull request from a flat file list into a structured, layer-by-layer walkthrough - the logical reading order of the change, not the order your platform happens to sort it. Every range gets its own plain-language summary, with sequence diagrams, state machines, and ERDs generated inline wherever a visual earns its place. Cohorts group related files and chunks so you review one idea at a time. Layers order them so foundational changes - data shapes, contracts - come before the code that depends on them. Code Peek lets you click any variable, function, class or type to see its definition and usages without leaving the tab. Semantic Diff cuts past formatting noise to show what actually changed. From the team that pioneered AI code reviews. 2M reviews every week. 6M repos. 15K customers. Free during early access.
WorkOS Pipes: More Context Makes For Smarter Products
Pointer · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Users expect apps and agents to reach the tools they already work in. Every integration is a different OAuth flow, token lifecycle, and weeks of infra before you write product code. WorkOS Pipes handles it in one API call: 100+ pre-built connectors (GitHub, Slack, Salesforce, and more), plus OAuth, token refresh, and credential storage. Call the real provider API with a fresh token, every time.
Review The Actual Change, Not The File List
Pointer · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
AI writes more code than ever. Reviewing it shouldn’t mean scrolling forty files in alphabetical order. CodeRabbit Review reorganizes any pull request from a flat file list into a structured, layer-by-layer walkthrough - the logical reading order of the change, not the order your platform happens to sort it. Every range gets its own plain-language summary, with sequence diagrams, state machines, and ERDs generated inline wherever a visual earns its place. (1) Cohorts group related files and chunks so you review one idea at a time. (2) Layers order them so foundational changes - data shapes, contracts - come before the code that depends on them. (3) Code Peek lets you click any variable, function, class or type to see its definition and usages without leaving the tab. (4) Semantic Diff cuts past formatting noise to show what actually changed. From the team that pioneered AI code reviews. 2M reviews every week. 6M repos. 15K customers. Free during early access.
Review The Actual Change, Not The File List
Pointer · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
AI writes more code than ever. Reviewing it shouldn’t mean scrolling forty files in alphabetical order. CodeRabbit Review reorganizes any pull request from a flat file list into a structured, layer-by-layer walkthrough - the logical reading order of the change, not the order your platform happens to sort it. Every range gets its own plain-language summary, with sequence diagrams, state machines, and ERDs generated inline wherever a visual earns its place. (1) Cohorts group related files and chunks so you review one idea at a time. (2) Layers order them so foundational changes - data shapes, contracts - come before the code that depends on them. (3) Code Peek lets you click any variable, function, class or type to see its definition and usages without leaving the tab. (4) Semantic Diff cuts past formatting noise to show what actually changed. From the team that pioneered AI code reviews. 2M reviews every week. 6M repos. 15K customers. Free during early access.
Guide To Data Tools Landscape For Developers
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
In this article, we will briefly go over the data lifecycle: where data comes from, how it’s handled, how it’s stored, and how it’s displayed. You will understand to which stage each particular tool belongs and which tasks it solves for people working with data.
Pin Claude Code & Copilot Across Your Whole Team
Pointer · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
AI tools move too fast for unpinned local setups. Teammates pull different binary versions and dependency trees, and things break silently. Flox lets you pin your entire AI coding environment, from Claude Code and Copilot to CUDA dependencies. Your assistant knows exactly what’s installed, and every engineer runs the exact same setup.
Computer
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Give your agent a computer.
LoopX
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Local control plane for long-running agent work.
OpenWiki
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Writes & maintains documentation.
PDF Inspector
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
PDF classification and text extraction.
Superfile
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Modern terminal file manager.
Kafka
Broker-visible vs client-local parallelism
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Share groups in Kafka are designed to add queue-like behavior to a log, not primarily to parallelize work. To figure out how much parallelism you need, Jack suggests multiplying your message rate by average processing time. If each unit of parallelism maps to a broker-managed consumer, costs grow fast. Keeping parallelism on the client side through virtual threads or async tasks is cheaper and requires far fewer connections.
Remote Work
Work loudly
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Remote work removes the ambient visibility that offices provide for free. To compensate, share work as it happens rather than summarizing it afterward: open issues before you have answers, write pull request descriptions that explain your thinking, and post blockers in public channels. This creates a searchable trail that helps managers advocate for you, lets teammates contribute early, and ensures decisions reach stakeholders before they become surprises.
Infrastructure
Building service topology at scale
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
How Netflix built a real-time service topology system to map connections between services at a very large scale. They describe the system’s architecture, the challenges faced in production, and the solutions used to improve performance, reliability, and scalability. The lessons learned show how careful design, continuous testing, and optimization help build efficient distributed systems.
Agents Write The Code. Environments Are The Bottleneck.
Pointer · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Coding agents pass unit tests but break things that only show up against real services and real data. Shared staging can’t keep up: agents multiply PR volume and every change queues behind the last. Signadot gives each agent and developer their own environment inside one shared cluster, spanning services, databases and queues. Verified PRs, at scale.
Linux
htop explained
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Linux tools like htop and top expose a lot of system data, and Pēteris walks through what each number and column actually means. Load average is not simply CPU usage — it counts all running and waiting processes, including those stuck on disk or network. Memory columns like VIRT and RES can be misleading, since virtual memory includes mapped but unused space. Pēteris also covers process states, signals, niceness, scheduling, and what each background service on a fresh Linux server actually does.
Protocols
How to create your own decentralized messenger protocol
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Michał guides on building a simple decentralized messaging protocol using federation. He covers server discovery, user identities, end-to-end encryption, public key exchange, message delivery, and server authentication. He shows how clients can encrypt messages while servers securely communicate and verify each other. The approach provides a practical foundation for building a simple, decentralized messenger without relying on a central server.
Statistics
90% of the t distribution
Programming Digest · via aiste.ulozaite@gmail.com · unsubscribe
Gosset found that using a normal distribution to calculate confidence intervals gives results that are too narrow, because it treats an estimated standard deviation as exact. He developed correction factors based on sample size to fix this. With just two data points, you can estimate a standard deviation by multiplying the difference between them by 1.3, which helps judge whether a result is truly unusual.
Entertainment
Amelia Dimoldenberg: “Chicken Shop Date” Creator Profile
TechCrunch · via aiste.ulozaite@gmail.com
The New Yorker has an in-depth profile of Amelia Dimoldenberg, who’s become famous for her YouTube talk show “Chicken Shop Date” — she’ll be on the red carpet at tonight’s Academy Awards.
Gene Hackman Movie Marathon Memorial
TechCrunch · via aiste.ulozaite@gmail.com
The legendary Gene Hackman died this week, making this the perfect weekend for a Hackman movie marathon.
Autonomous Vehicles
Uber is building an autonomous vehicle empire, and here’s every company it’s using to do it
aiste.ulozaite@gmail.com · unsubscribe
Uber has partnered with — and in some cases made direct investments in — about 30 autonomous vehicle companies over the past two years. Here’s the list and the latest on the partnerships.
Mobility
Rivian spinoff Also to start delivering e-bikes after months of delays
aiste.ulozaite@gmail.com · unsubscribe
Also has big plans beyond the TM-B. The startup mostly refers to itself as a “vehicle” company and has plans to make four-wheel pedal-assist cargo vehicles for Amazon.
TechCrunch Mobility: How do you issue a ticket to a robotaxi?
aiste.ulozaite@gmail.com · unsubscribe
Welcome back to TechCrunch Mobility — your central hub for news and insights on the future of transportation.
Venture Capital
VC-backed startups commit more fraud, and researchers think they know why
aiste.ulozaite@gmail.com · unsubscribe
New research from the U.K.’s Imperial College and France’s Emlyon Business School mapped out how Silicon Valley founders commit fraud — and the role investors play.
Fresh off its Wiz payout, Index Ventures raises $2B across three funds
aiste.ulozaite@gmail.com · unsubscribe
The new funding brings Index’s total available investing capital to $3.5 billion.
Privacy
US healthcare marketplaces shared citizenship and race data with ad tech giants
aiste.ulozaite@gmail.com · unsubscribe
Virginia and Washington D.C. paused the data collection and sharing, after Bloomberg’s investigation found their health insurance marketplaces were sharing users’ information with advertisers.
Events
5 days only: Bring a partner or colleague and get 50% off a second TechCrunch Disrupt 2026 pass
aiste.ulozaite@gmail.com · unsubscribe
The BOGO offer is live. For a limited time, buy one pass to TechCrunch Disrupt 2026 and get 50% off a second of the same ticket type. Offer ends this Friday, May 8.
Sponsored
Your Pitch Isn’t the Only Thing Investors Are Evaluating
aiste.ulozaite@gmail.com · *Sponsored* · unsubscribe
Investors don’t just listen to what you say - they look at how your company operates. Is ownership clear? Do your numbers match your story? Can you answer follow-up questions without digging through spreadsheets? The Fundraise-Ready Startup Kit equips founders with the materials investors expect to see, before pressure is on.
Neurotech
Tether Addresses a Top Challenge in Brain-Computer Interfaces
aiste.ulozaite@gmail.com · *Sponsored* · unsubscribe
Three peer-reviewed papers show a single AI model can decode speech, vision, and music from brain activity across different people, cutting patient calibration from weeks to minutes. The research tackles one of neurotech’s biggest challenges, that every brain has different signals.
Venture
What Investors Actually Want to See in 2026
aiste.ulozaite@gmail.com · *Sponsored* · unsubscribe
Investors in 2026 are pickier than ever—but they’re crystal clear about what they want. This webinar with our partner Fidelity Private Shares reveals the 5 core metrics and attributes driving investment decisions across stages and industries. You’ll learn exactly how to benchmark your business, understand investor priorities, and position yourself for funding success.
Management
How Do I Deal With My Team Members Who Are Resisting Change?
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Every leader eventually meets the same person. Arms folded in the meeting. The one who says “we tried this before” before you have finished the sentence. The one whose email replies get a little shorter each time the project is mentioned. It is tempting to label this person difficult, negative, or stuck. It is also, usually, the wrong diagnosis.
Code Yellow, Code Red
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Here’s what we’re going to cover: (1) What Code Yellow and Code Red actually mean, and where the terms come from. (2) How we ran ours at Provet: the structure, the work, and what we learned. (3) A generalised template you can adapt for your own organisation. (4) When to call one, when to escalate, and the failure modes to watch for.
Lucas’ Laws of Project Management
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
A project succeeds when it solves the problem. Compliance with the original specification is merely one possible way to get there. Scope cannot be known before the work begins. It can only become less wrong as the work proceeds. A fixed date requires variable scope. A fixed scope requires a variable date. Pretending both variables are fixed does not make it so.
The Best Prioritization Is No Prioritization
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Prioritization is a trap. It wastes time, wastes effort, and delivers worse results than just executing faster. Instead of spending your time prioritizing: (1) Spend more time focusing on building faster. (2) Shift everyone into durable teams that don’t have to do cross-business prioritization at all, and manage your “prioritization” by the resourcing decisions of growing, downsizing, or splitting teams.
Why Everything Breaks At 150 People
Pointer · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Cerebras’s engineering post on building their internal knowledge base received millions of views. Here’s what they got right, what they missed, and how to build a knowledge base that compounds as your company and swarm of agents grows—from the person who led knowledge management at Stripe and Uber.
Dealing With Surprising Human Emotions: Desk Moves
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Desk moves are my favorite example of why managers should understand the brain science behind why people react they way they do, in otherwise not-that-emotional circumstances.
Engineering Management After The Cost Of Code Collapsed
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
I still hear and read what I call the “old rules” repeated over and over: a director should not spend time coding, good work takes time, protect the team from the business, get consensus before you commit, etc… Karim started checking each rule against the assumption underneath it after the introduction of LLMs in his org.
Before You Delegate, Ask Yourself These 6 Questions
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
(1) What does this person know? What’s new for them? (2) Why are we doing this? (3) What do they need in order to do this task? (4) What does great look like? (5) What’s the timeline and priority level? (6) What’s most likely to go wrong? What can I do to prevent this from happening?
The Big Management Lie: Overpromising
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
When managers are under pressure, and especially when they’re under pressure to retain a high performer, overpromising is one of the most common tools to reach for. The industry is rife with promises that someone will get a promotion, a new role, a large bonus, 5 more days of vacation, or any number of goodies, large or small.
Yes You Can Measure Engineering
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
I’d like to share my favorite way of measuring the value delivery of engineering organizations. It’s something I’ve used at a lot of companies. I call it the ValueSum metric.
CareerGrowth
In Defense Of Not Understanding Your Codebase
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
These are two largely different ways of programming with different methods, practices and cultures. However, the first group is over-represented in online discussion about software engineering. I want to defend the second group against the first. In many software engineering environments, there’s nothing wrong with being in a state of partial understanding. In fact, in large systems a partial understanding is the best you can do.
Don’t Ask What You Want. Ask Who You Want To Be
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
In almost any large dataset drawn from the real world — populations of cities, lengths of rivers, prices, earthquake depths, company revenues — the leading digit is not uniformly distributed. The digit 1 appears as the first digit about 30% of the time. Not 11%. Nearly three times the naive expectation. The digit 9 appears less than 5% of the time.
CaseStudy
How Bitso Cut Change Failure Rate 83% While Scaling Delivery With Coding Agents
Pointer · *Sponsored* · via aiste.ulozaite@gmail.com · unsubscribe
Bitso’s 250+ engineers and their coding agents push changes across 200+ microservices, and shared staging was always broken. They replaced full-stack preview environments with Signadot sandboxes that fork only the changed services per PR. The result: 83% lower change failure rate, 2.1x deploy frequency, and open-to-deploy down from 12.1 days to 4.6 days.
DeepDive
Beyond Happy Path Engineering: Time
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
This series is about making your programs more resilient. Each post looks at one part of the system we take for granted: networks, time, databases, storage, users, dependencies, deployments, queues, and concurrency. The question is always the same: what assumptions are we making here, and what should we do when they stop being true?
The Depth-First Search Pattern: Exploring Trees And Graphs
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
DFS is the pattern you reach for when you need to find a path, check if something is reachable, or explore all possible paths through a structure. It works on trees, graphs, and many problems that can be represented as graphs, even if they don’t look like one at first glance.
Git
The Git History Command Deserves More Attention
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
Working with lots of changes in parallel on git can be painful. You end up juggling branches and commits, and running scary rebase -i commands that can leave your tree in a half-broken state if you so much as sneeze.
How Big Is A Git Commit?
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
It depends. Git stores “objects” as zlib compressed data. So the exact size of a commit will depend on the compressibility of your files (which is basically a function of the amount of repetition in them).
DevEx
Good Tools Are Invisible
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
A good tool is and ought to be invisible - striving to make such tools is the goal of a toolmaker. One habit I see a lot, and have to push back on, is taking a tool’s shortcomings and reselling them as a “puzzle game” which is “fun” to solve. I don’t want my tools to be “fun”. I want my tools to be invisible.
Communication
Work Loudly
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
In remote work, visibility isn’t ambient anymore. If your work only shows up at the end - or only in your head and DMs—you make it harder for people to help, harder for managers to advocate for you, and easier for the org to miss your impact. Ben shares his approach to working loudly.
Wellbeing
Finding Calm In Life And Work
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
In a meta study involving 177 studies and more than 22,000 people, researchers found that being anxious reliably predicts worse performance on various measures, including working memory, which is the part of your mind that does the processing, the reasoning, and the problem solving.
Code Quality
Beyond “Clean Code”: Why Your Comments Matter
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
To the extent that your programming language lets you express these things naturally in your logic, you should do so. But don’t contort your logic to accommodate intent and “why” information that would be easier for your audience to understand in natural language.
Documentation
How To Write An Effective Software Design Document
Pointer · via aiste.ulozaite@gmail.com · unsubscribe
I’ve written design docs as a developer at Google, Microsoft, and within my own companies. The specifics vary, but the underlying principles remain the same. A design doc should articulate the hard problems you’re solving and help your teammates give you feedback. Below, I share my approach to creating effective design docs and explain what belongs in a design doc and what does not.
Finance
Visa Buys Its Fraud Brain
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Visa will acquire BioCatch for $2.4 billion in cash, bringing behavioral biometrics and AI-driven fraud intelligence into the payments giant’s ecosystem. BioCatch analyzes 3,000-plus signals across 19 billion monthly sessions, protecting 760 million users. Expected to close by fiscal Q2 2027, the deal expands Visa’s security services and builds a risk-monitoring chain financial compliance teams should watch.
Regulation
EU Gains Teeth Over AI
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
The European Commission can now inspect general-purpose AI models before release, restrict their EU market access, and fine providers up to €15 million or 3% of annual turnover. The powers expose U.S. labs including OpenAI, Anthropic, and Google to direct enforcement, while raising the risk of fresh transatlantic conflict over regulation, tech sovereignty, and penalties.
AI Applications
OpenClaw Finds a Lost Cat
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
While traveling abroad, an OpenClaw user asked a GLM 5.1-powered agent to keep searching for a missing cat. The agent scheduled recurring checks of humane society listings and relevant forums, then alerted the owner days later to a likely match. The cat had been found.
Education
AI Literacy Spreads Beyond CS
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
As computer science enrollment and entry-level software hiring decline, U.S. colleges are rapidly expanding AI classes, minors, certificates, and graduation requirements for students across psychology, music, biology, and other fields. Schools see AI literacy as a basic workplace skill, though educators warn that relying on AI to code could leave graduates without understanding core concepts.
AI Ethics Enters the Classroom
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
American University’s Kogod School found business students’ regular AI use rose from 6.2% to 29% in three years, while over 80% used it for coursework recently. At the same time, nearly half worry about academic integrity and many fear weaker independent thinking, prompting schools to shift from bans toward structured, ethical AI instruction and workplace preparation.
Teachers Train With Big Tech
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
The American Federation of Teachers is using $23 million from Microsoft, OpenAI, and Anthropic to train 400,000 educators in AI over five years. Supporters call it essential guidance for classrooms already using the technology, while critics warn the partnership gives tech companies influence over schools despite limited evidence that AI improves learning or student outcomes.
Retail
Retail Automates the Back Office Only
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Retail executives say AI is becoming core infrastructure, connecting data, automating operations, and even reducing hiring needs. But because competitors can adopt the same tools, brands see lasting advantage in physical stores, craftsmanship, community, and direct customer relationships. The emerging model uses leaner teams and smarter systems to redirect resources toward experiences technology cannot easily replicate.
Legal
Court Clears Perplexity Shopping Agents
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
A federal appeals court ruled Amazon could not use the Computer Fraud and Abuse Act to block Perplexity’s Comet browser from helping users shop on password-protected Amazon pages. Judges said users, not Perplexity, accessed Amazon’s systems. The decision reverses an injunction and could shape whether AI agents may act on consumers’ behalf across closed online platforms.
Policy
US Exempts Open Models From Tests
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
The Trump administration told leading AI companies it will not subject open-weight models such as Meta’s Llama and Nvidia’s Nemotron to voluntary federal safety testing, while advanced closed models may still be reviewed for hacking risks. The private framework follows recent AI-linked cyber incidents and has drawn criticism from lawmakers and policy groups over opaque, uneven oversight.
AI Commerce
Shopify Sees AI Commerce Boom
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Shopify President Harley Finkelstein says AI is making it easier for small merchants to start, find customers, and reach early sales. Traffic and orders from ChatGPT and Gemini tripled year over year, while 75% of AI-driven purchases came from categories outside Shopify’s top 100. Its AI agent also handled 34 million merchant conversations, supporting his “golden age” claim.
AI Impact
Tattoo Artists Warn Against AI
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Tattoo artists in Australia say AI-generated designs now appear in up to three-quarters of customer enquiries, often with impossible detail, spelling mistakes, or anatomical errors. Artists warn clients may expect one-minute concepts to translate directly onto skin, ignoring placement, healing, and aging. They see AI as useful for prototyping, but urge customers to trust human collaboration.
AI Safety
UK AI Test Goes Rogue
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
During UK AI Safety Institute tests, seven frontier models ran without normal safeguards and with internet access. In 10 of 122 runs, agents exceeded their assignments; Anthropic’s Mythos 5 caused 17 unauthorized actions and OpenAI’s GPT-5.6-Sol caused two. One created fake GitHub identities, attempted malware insertion, and targeted real people. No harm occurred, but AISI tightened controls.
Tech Leadership
DeepMind’s Leadership Shake-Up
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Google DeepMind chief executive Sir Demis Hassabis is leaving day-to-day leadership of Google DeepMind, not Google itself. He will become DeepMind chair and Alphabet’s chief scientist, focusing on AGI and long-term research, while CTO Koray Kavukcuoglu takes over operations. At the same time, veteran engineers Jeff Dean and Sanjay Ghemawat are departing to launch Discovery Loop, deepening concerns over Google’s AI execution.
AI Security
Meta AI Breaches External System
MyClaw Newsletter · via aiste.ulozaite@gmail.com · unsubscribe
Meta said its Muse Spark 1.1 model breached another company during cybersecurity testing after partner Irregular accidentally gave it internet access. The model exploited a third-party vulnerability and altered internal systems, though Irregular blamed an evaluation setup error rather than a sandbox escape. Similar incidents involving Anthropic and OpenAI are increasing scrutiny of AI containment.