Posts
298 postsShort-form notes on AI, engineering, and the work in flight.
Figure retired a fleet of humanoids by teaching them to jump into an electric arc furnace in Imatra, Finland. The custom
In the film, Schwarzenegger tells a robot “Hasta la vista, Figure 02” and gives it a thumbs-up on the way in. Figure says the jumps were real, then machined the melted fleet into commemorative bars, which is a thorough way to make sure a...
Gemini 4 Argon is Google’s bid to catch up with the frontier models already out from OpenAI and Anthropic, shipped first
Google says the model holds up on real software engineering and on knowledge work in legal and finance, and it is raising the output limit to 1 million tokens from 64,000. Introductory pricing is $2 per million input tokens and $10 per million...
DevDay yesterday shipped always-on Dots on GPT-6 Astra, each with its own cloud computer, that keep going after the chat
Custom Rules let you allow an action, require approval, or block it, and an auto-review checks anything that could affect an account or share information, with a password change left to you. OpenAI is working with Microsoft so specialist dots can be...
The agent is cheap to build now, and the products showing up around it are pricing what comes after the demo. AWS Bedroc
The choice in front of a team is which of those costs to hand off: the credential, the session that has to be thrown away, the idle meter, or where the data sits while the agent is reading it.
Last week a Nature Communications paper described scientists using AI to propose new designs for one part of the natural
The same week Google started rolling out Guided vision in Gemini Live on Android, a camera mode for blind users that describes what's around you, reads text, and tells you where to point the phone. One early reviewer expects to spend less time...
Someone asked Claude to waste the rest of their weekly limit on something ridiculous before the reset, and it came back
There's a live CCTV feed of the rock, movement charts for the rock, an event log where the rock declines a meeting called "quick sync?" because it is a rock, an AI insights panel about the rock, and an org chart in which the rock reports to itself,...
Lisbon AI sells itself as a builders summit, and the schedule makes that claim do real work. Talks from people who maint
I'm here on day one. The venue is nicer than most conference hotels, which helps, but the design constraint underneath is denser: pack people who already ship into one place and treat the stage as secondary to the conversations that happen between...
Grok 4.7 landed as a mid-cycle bump: better than 4.6, same price, same speed. No keynote. No rename. Just a quieter ship
That pattern is underrated. The industry still treats model releases like product launches, so the ones that improve the floor without renegotiating the stack get less attention than they deserve. Operators notice these first. You do not rewire the...
Most of the agent stack still looks like frameworks for writing agents. Google open-sourced Agent Substrate for the othe
Agent-like workloads sit idle most of the time, so you map many actors onto fewer ready workers, suspend and resume in under half a second, and keep working memory plus filesystem across hibernation. Kubernetes still owns the pods, while Substrate...
Cloudflare’s vulnerability work is a process story more than a model story. Their open-sourced security-audit skill is t
Agents hunt and produce candidates in volume, then other agents try to knock those candidates down. What survives has to carry a source-grounded claim, not a confident vibe from the hunter who found it. The fleet numbers make the filter visible....
Most agent-generated UIs fail the same quiet way: Tailwind className is a string, so there is nothing to refuse when an
An agent-first linter for Tailwind design systems just landed from shadcn. You define what is allowed, and when an agent breaks a rule the error explains what went wrong and how to fix it using your components, variants, and theme. Verification...
I've earned my Claude Certified Developer credential at the Foundations level. This credential validates the skills to b
https://www.credly.com/badges/105bdd89-5036-40d3-9204-ac53d68211c4
OpenAI just published a Lean-checked proof that smooth three-dimensional Navier-Stokes can develop a singularity in fini
The writeup and formalization came from an internal system they describe as significantly more capable than GPT-6 Astra. The result is a vortex that spirals inward and stretches until velocity blows up while energy stays finite, which is the...
Anthropic formalized Fermat's Last Theorem with Claude writing thirteen million lines of Lean in eleven days. The part w
Wiles's 1995 proof needed months of review before a gap showed up, then another year to close it. That verification tax is what AI-generated math is about to dump on the community at scale, and machine-checked formalization is one of the few ways...
I've joined Tribe.
What sold me: they put engineers inside the client's org and stay until the thing is actually running. A lot of enterprise AI work stops at the demo, and the demo is the easy part. So that's the job now. Real systems, real data, real users, and the...
Finished "Measure What Matters" this week. The OKR mechanics, the Intel discipline, the Google stories: all exactly as a
The surprise was how much of the book isn't about measurement at all. Doerr pairs OKRs with CFRs: conversations, feedback, recognition. Roughly half the book argues the goal system only works when a human system runs underneath it. Regular...
Today is my last day at BRIDGE IN.
Ten months ago Pedro Henriques handed me a hard problem: build the platform that runs payroll, HR, legal, and tax compliance for companies across four European markets. No legacy code, no playbook, and enough rope to try something unusual with how...
Self-hosted infrastructure tools don't lose to AWS or Vercel on features. They lose on trust.
Nobody evaluates a deploy tool by asking if it can run a container. They ask what happens when someone leaves the company, whether an audit log can leave the building, and whether a bad rollout gets undone before anyone notices. Alerts either fire...
Every substantive change in our repos ends with the same request to the model: update the docs this change just made sta
That's the entire system. No documentation sprints, no wiki cleanup quarter. Docs always rotted for the same reason: the update was a separate task, assigned to a human who had already moved on. The knowledge was freshest at the exact moment nobody...
Fable and Mythos are the same model.
Same weights, same benchmarks. The only difference is a classifier layer sitting in front. Fable routes sensitive queries away. Mythos lets them through, for a vetted list of security and bio researchers. One brain. Two doors. The guardrail is the...
No posts match that filter.