Yes. For example, postgres has a native uuid type, and it's 16 bytes. Every varchar(36) uuid column I've seen in review came from an ORM default nobody questioned.
If you keep correcting the same mistake, make it a check.
@poteto of @spacexai on turning corrections into deterministic checks, and keeping the codebase worth copying.
From AI Engineer Code Summit.
In "stuff that @Steve_Yegge said 9 months ago, when few believed it would be true, but now it's more visible"
The "vampiric" effect of running AI agents, and how it drains people
I don't know how he does it, but he rarely misses
Full: https://t.co/fYR25YHscJ
VS Code made the Copilot harness the default for agent sessions; the Local harness goes away in Feb 2027. Microsoft's evals, same models: hangs and crashes 3.1% to 0.2%, completion 54.8% to 61.2%. The harness is the product now. Check your hooks.
https://t.co/T0hab2sQNb
Some of you have been asking very politely. Some of you have just been asking.
GPT-6.2 Astra and Sol are now available.
Super efficient and built to perform where it counts—the real world.
https://t.co/qr2iP0GpfX
6. Review it like code. A rule earns its line when the agent repeats a mistake. Contradicting rules get one deleted. Monorepo? Nest files: the closest AGENTS.md to the edited file wins.
Agents don't read minds. They read files.
https://t.co/YWaWFvGnqV
Over 60k open-source projects now ship an AGENTS.md, the README for coding agents that Cursor, Copilot's coding agent, Junie and others support.
Most of the value is in six boring sections. Here's how I write mine so the agent stops guessing.
5. A definition of done. Which checks must pass and what proof to paste back: test output, the command it ran, a screenshot for UI. Without it, "done" means the agent stopped typing.
Most AI dev tools optimize for writing more code faster. One built around understanding the code is the rarer bet, and probably the one that ages better.
We built a tool for our own internal team at AnswerAI to let us learn, explore, & develop mastery with the help of AI: SolveIt.
Today, we’re releasing it to the public for the 1st time.
https://t.co/PeGT8WWcxF
Microsoft and Berkeley open-sourced TokenEconomizer for Claude Code, Codex and Copilot CLI: learns repo quirks, prunes unused tools, breaks failure loops. 18.4-30.9% lower mean cost on SWE-Bench. You were paying an agent to rediscover pytest is missing.
https://t.co/mLwCx44oEZ
GitHub's Copilot app now lists code scanning alerts by severity, with Explain, Fix and /security-review before the PR. Fix only edits your local copy. No commit, no push. Correct call. The alert stays open until you push and the scanner agrees.
🔒 Check your repo's code scanning alerts without leaving the GitHub Copilot app.
With the new GitHub security extension, you can ask Copilot to find, explain, and fix (or dismiss!) an alert.
Now in public preview ⬇️
https://t.co/RE7rcSWtle
Make inference 10x faster. Now the tool calls are the bottleneck.
@nserrino of @gimletlabs walks through why faster models make the rest of your agent stack matter more.
From AI Engineer Code Summit.
Revolut shared the postmortem of their global cards outage of 5 hours. Some interesting (alarming?) notes:
1. They had no monitoring on database nodes going down (!!)
2. Across 3 regions, they ran 5 nodes, with #1. So a single node going down meant no zonal resillience (cont'd)
Fireworks exposed a key Cursor issued it. The attacker copied 2025 agent traces, some with unredacted credentials, used those on Cursor's SaaS tools, then parked a fake Claude Code build on its extension cache ~11h. Agent transcripts are secrets files.
We’ve published an update on our recent security incident. See a detailed overview of what happened and our remediation steps here: https://t.co/2lf6lRJPlL
This is the right instinct. The agent shouldn't own the control flow, it should fill in transitions you already defined. Reviewing a diagram of states beats reviewing 2,000 lines of almost-right diff.
So one thing I have been struggling with in this agent era is having what I think should be done and the agent doing it. There is so much control that is lost.
So here my current idea. The state-machine that is my program is now visualized! (moar posts below)