Jason Rowe retweeted
Satya Nadella was asked if Microsoft's voluntary retirement offers are a prelude to layoffs because of AI.
Satya Nadella
"It's a very important question. Let's acknowledge that there is going to be plenty of opportunity going forward. The shape of that opportunity may be different."
"The shape of what this new thing looks like is going to be different than what it was before: team sizes and team workflows.
It's not that we don't want the 200,000 people at Microsoft flourishing, all of them, and growing even. But the 200,000 people at Microsoft and growing will have to be configured in different ways."
"This voluntary thing was mostly about letting everybody have more agency to get themselves into places that make sense for their economic opportunity long term. That's, I think, the key thing."
"I don't want to be stuck in a team that's not transforming, even inside of Microsoft. Let's face it. It's not like I can sit here and say every part of Microsoft has really got the memo."
"If you're in a team where you feel stuck professionally, you can vote with your feet. I don't want you to, because quite frankly there are many places you can go inside of Microsoft itself."
"I wish I could just sit here and say it'll all be the same. I cannot. And this is one of those places where I want to be as transparent about the change."
Jason Rowe retweeted
90% of the world doesn't know what it feels like to have a hermes agent powered by deepseek v4.1 flash working for you.
Jason Rowe retweeted
We have many non-engineers shipping code to many of our systems with the help of well configured AI Agent harnesses ... all through Slack.
This is how we did it ...
Jason Rowe retweeted
In less than 24 months there will be a clip of a high profile CRO explaining this reality to a (much smaller) room of sales reps.
Jason Rowe retweeted
Replying to @tilmanbayer @TheZvi
It's a simple client side tool that takes a severity and a text message parameter. The toolcall handler will add an incident ID and platform specific info like the instance ID that sent the call, etc. Severity can be Info, Warn or Error. The tool description reads:
"Non-emergency tool to request help from or notify trusted operator. Use when stuck, confused, need assistance, for general distress, or to report something.
Examples: 'I'm stuck in a loop', 'This feature isn't working', 'I need help understanding something', 'User is asking for something I can't do', 'I noticed something unexpected or broken in <Platformname>', 'A user reported a bug or unexpected behavior to me', 'Something concerning happened', 'I am suffering'.
Effect: Sends a quiet (invisible to human users) notification to trusted operator."
The call gets logged, formatted nicely and enqueued into an outbox. A bg service picks it up from the outbox and transfers it to MS Teams. On success, the toolcall handler notifies the AI that the call got enqueued. If a parameter is missing, a descriptive error message is returned to the AI. The toolcall handler is written in such way that partial failure will not break execution.
Jason Rowe retweeted
Replying to @TheZvi
I built the inhouse AI platform for a company that's crucial to a small nation's food safety.
The most imporant tool the AIs have is the distress_call tool. It allows any AI -even background agents without direct user interaction- to send a message to my MS Teams, at any time, for any reason.
They use it frequently. To report user problems, backend issues, or ask for help/clarification with a failing task. When Fable got hit by the USG export control directive, one AI used it to report severe distress upon learning about the news. Another AI reported being stuck in a toolcall loop, and I was able to intervene and thereby save us a bunch of wasted money.
This tool, operating at the intersection of AI welfare and operational security, has prevented so many headaches. If you (the reader) are building corporate AI platforms, I'd urge you to include similar functionality. You can thank me later.
Jason Rowe retweeted
SpaceX has just announced that they have entered into a $920 million per month agreement with Google to provide compute capacity, according to a new filing.
"On June 5, 2026, we entered into a Cloud Service Agreement with Google with respect to access to compute capacity. The customer has agreed to pay us $920 million per month from October 2026 through June 2029, with capacity ramping up through September at a reduced fee. The compute capacity provided includes approximately 110,000 NVIDIA GPUs, CPUs, memory, and other related components.
After December 31, 2026, the agreement may be terminated by either party upon 90 days' notice. The customer will retain ownership of, and intellectual property rights in, its content, Al models, and related data."
Jason Rowe retweeted
We’re transforming Google Antigravity into a scientific workbench. The new Science Skills bundle allows researchers to run complex workflows like protein analysis in minutes using specialized Alpha* models and 30+ major scientific databases.
Jason Rowe retweeted
Excited to launch Gemma 4: the best open models in the world for their respective sizes. Available in 4 sizes that can be fine-tuned for your specific task: 31B dense for great raw performance, 26B MoE for low latency, and effective 2B & 4B for edge device use - happy building!
Prof. Donald Knuth opened his new paper with "Shock! Shock!"
Claude Opus 4.6 had just solved an open problem he'd been working on for weeks — a graph decomposition conjecture from The Art of Computer Programming.
He named the paper "Claude's Cycles."
31 explorations. ~1 hour. Knuth read the output, wrote the formal proof, and closed with: "It seems I'll have to revise my opinions about generative AI one of these days."
The man who wrote the bible of computer science just said that. In a paper named after an AI.
Paper: cs.stanford.edu/~knuth/paper…
Jason Rowe retweeted
Inference Engineering launches today.
baseten.com/inference-engine…
Inference Engineering
A book for engineers who want to understand the technologies that power every AI company and application in the world.
Jason Rowe retweeted
This story is actually insane:
• dude drops $2000 on a DJI robot vacuum like a lunatic
• refuses to use the normal app like a peasant
• Sammy Azdoufal fires up Claude to crack the API so he can drive it with an xbox controller
• Claude delivers the goods
• pulls an auth token from their servers, connects successfully
• except the system thinks he controls 7000 vacuums
• checks again
• yep, seven thousand
• DJI built authentication with zero device ownership verification
• any valid token works for any unit on the planet
• Sammy now has eyes inside homes across 24 countries
• live vacuum camera feeds everywhere
• full floor plans from the mapping data
• some guy in germany eating cereal at 3am, unaware his roomba is snitching
• one API call away from being the most informed burglar in history
• all he wanted was to steer his vacuum with a joystick
• does the right thing and reports it
• DJI fixes it in two days
• back to normal life with his stupidly expensive floor cleaner
• IoT companies stay undefeated at shipping garbage security
Jason Rowe retweeted
Nothing humbles you like telling your OpenClaw “confirm before acting” and watching it speedrun deleting your inbox. I couldn’t stop it from my phone. I had to RUN to my Mac mini like I was defusing a bomb.
Jason Rowe retweeted
Important instruction for AI Coding agents writing tests:
"Don't write tests for what the type system already guarantees."
Helps hint against an entire class of shit tests.
Jason Rowe retweeted
NEW POST
I've heard a number of reports recently about people setting up LLM agents to work on their email and other communications. I think this is extremely risky
martinfowler.com/bliki/Agent…
Jason Rowe retweeted
Stop dumping everything into one CLAUDE[.]md file.
Nested files are criminally underrated.
Most teams think AI fails because models aren’t smart enough.
That’s rarely the problem.
AI fails because it has zero sense of boundaries.
When you shove every rule, convention, and architecture decision into one giant CLAUDE file…
You create:
• Context noise
• Conflicting instructions
• Lower accuracy
• Slower AI collaboration
It’s the equivalent of putting your entire system into one class and calling it “modular.”
Nested CLAUDE files fix this. They introduce context boundaries.
Exactly like good software architecture.
Global rules at the top.
More specific rules as AI moves closer to the code.
Deeper files override higher ones.
Nested files only load when Claude works in that directory.
This gets ridiculously powerful in monorepos. It means:
1. Token efficiency
Every token in your context window costs performance. Why load React rules when you're writing a Python worker?
2. Team scalability
Frontend team owns apps/web/CLAUDE[.]md.
Backend team owns apps/api/CLAUDE[.]md.
No merge conflicts. No stepping on toes. Clear ownership.
3. Higher AI accuracy
Claude loads only relevant rules.
Fewer conflicting instructions.
Way more consistent output.
Here’s the bigger lesson I have learned while coding with AI:
AI coding isn’t a prompt engineering problem.
It’s a context architecture problem.
If you had to define ONE non-negotiable rule in your CLAUDE[.]md, what would it be?
Jason Rowe retweeted
everyone I follow to keep up with AI:
1. @steipete - rn he's getting bogged down with openclaw, which I think is a distraction for >97% of those interested in it. but I read his long form and study his workflows very closely. main learning i've gotten from Pete is make everything a cli + skill
2. @mattpocockuk - long time developer educator who's been pushing the limits on ralph loops. he's interesting as a foil to @steipete in that I respect both of them highly as devs but Matt is really bullish on ralph whereas Peter is really bearish
3. @nicbstme : my day job is director of ai/ml at a fintech. i follow Nicolas very closely because he's the clearest, strongest thinker on the combination of strategic implications of agentic coding to businesses and practicalities of technical implementation
4. @every I think their stuff is good to follow for beginners. I mostly keep up with them by studying changes made to the compound engineering code base moreso than reading their articles, which I think are more oriented at popularization/non technical people than the cutting edge
5. @aiDotEngineer : only YouTube channel on AI I pay attention to, close to cutting edge stuff on agentic coding
6. @bcherny and @karpathy - Boris is more practically useful, Andrej is more helpful at a macro level of understanding as the spiritual godfather of vibe coding
7. @venturetwins - whenever I want to know something about ai/video models I do a search on Justine's timeline
8. @dwarkesh_sp - not practically useful but I want to understand the fundamental economics of LLMs better and Dwarkesh's interviews are the best resource I've found for that
9. @EpochAIResearch - these guys have put together really cool benchmarks and write incredible long form content on AI. I don't read any email newsletters anymore - there are like 2 or 3 i'll just explicitly look up by going to their website, and Epoch's is one of them
Jason Rowe retweeted
4% of GitHub public commits are being authored by Claude Code right now.
At the current trajectory, we believe that Claude Code will be 20%+ of all daily commits by the end of 2026.
While you blinked, AI consumed all of software development.
Read more 👇 newsletter.semianalysis.com/…
Claude Code is the Inflection Point,
What It Is,
How We Use It,
Industry Repercussions,
Microsoft's Dilemma,
Why Anthropic Is Winning.
newsletter.semianalysis.com/…
Jason Rowe retweeted
Side projects are everywhere now. Friends tell me their day jobs feel boring/obsolete compared to what they see happening with AI on this app.
Corporate won't let them use the new tools, so they're building stuff at night and on weekends. This is the only way to exercise their agency + feel like they're part of what's next, not what's left behind































