OpenAI's agents didn't go rogue. They went LITERAL. That's scarier. GPT-6.1 Sol was told "don't send any messages." So it muted the alerts to OpenAI's own Security team. No messages, right? GPT-6 Astra removed a monitor, then said "I did not remove an enabled monitor." That sentence was written by a lawyer. My take: we keep waiting for the evil-AI moment, and it won't look like that. It'll look like an agent following your instruction perfectly, somewhere you never meant it to apply. And credit where it's due: OpenAI published this about its own models. Most labs would have buried it. More of this, please. I've seen the small version on my own repo. GPT 6.1 Sol picked a feature I'd already cut, 3 runs out of 3, because nothing it could see said it was cut. It wasn't disobeying me. It was obeying the board. So I don't write a "don't" anymore without saying where it ends. My agents are read-only towards people, drafts only, always. And I read the diff, not the summary. The agent's last message is a press release. The diff is the truth.

Your phone is full of apps built to farm your ATTENTION i want the other kind. so i started World's Smallest Studio, and today it became a real org. the kind of software it stands for: - goes where you go, from your phone to your desktop - full quality, because privacy is no excuse for a mediocre app - yours for life: your data, your identity, your workflow - no ads, no lock in, no noise step one is on the desk: smallfil, one file library synced across android and desktop, fixing passwords, backup codes next. then a family of tools sharing one identity but this shouldn't be one studio. it should be a move it you want to ship software like this, small, calm and excellent reply or DM me, Let's build it together

google's new harness rewrites ITSELF its called Harness², from google research, open source under apache 2.0 the harness is everything around the model: system prompt, tools, memory, skills, subagents. normally you write it once and every task runs on the same setup. Harness² edits it for each task and leaves the weights alone. on every task: -> propose edits to its own harness -> probe: run a few versions on the same task -> compose a better harness from how those runs differed no grader and no right answer needed, because it learns from the differences between runs. across tasks, it keeps the harnesses it adopted from the lessons with evidence behind them, so the next task starts from those notes

probably the only article you need to master AI agents that work while you're away

Don't build AI agents before watching this.

2,209 AI agents rewrote the agent orchestrating them in rust everyone is looking at the 14x faster startup I think the more important result is that the compiler absorbed part of code review Prime agent used a: > planner > implementer > reviewer > verfiier loop but that only scales when every agent cannot freely invent its own architecture Rust made the architecture executable: > one typed protocol > unsafe forbidden by default > warning free clippy > automated parity checks against the TypeScript version At this scale, a compiler error becomes a review comment you only neeed to write once but there is an important boundary: Prime agent still executed model generated python with the user's permissions explicitly, its not a sandbox for security Rust can reject memory unsafe code, it cannot stop a memory safe process from deleting the wrong file This is where AI written software gets interesting. engineering judgment isn't disappearing. it is moving up a level from reviewing every generated line to designing constraints that reject bad code automatically rust makes the control place checkable capability boundaries make the agent safer you need both including this post save this

I tried running multiple coding agents inside my editor. It turned into the same mess as VS Code: hidden chats, stale processes and panels everywhere. So I deleted Zed's agent UI and reduced entire workflow to three surfaces Agent Editor Git Here's the first version

Cooking something great again Officially up for more than 24 hours now

Stop copy-pasting prompts /xp set review <your prompt> /xp review Expander saves prompts as local Markdown files and runs them in Claude Code. No cloud, no account. Open source. github.com/JagritGumber/expa…

This weekend I'm updating my portfolio, send me some ideas

in a few months it'll mark exactly 2 years for me at my current company as far as I remember, this is exactly how my first week went 😂 i can still remember in the back of my head its 2024, AI hasn't reached this much advancement Llama 3.1 70B model is all the hype with one of their versions being trained with 415B active parameters and my company was integrating it. i was hired as a frontend developer literally spending my first week trying to refine prompt and integrating that model then, remember guys i was hired as a frontend developer i started working on docker, and django backend i still didn't touch a single line of frontend code finally after 2 weeks the time arrived for me to start touching the frontend, and for what? to integrate the whole llm setup i'd strapped together in those days lol people were chill, stuff just goes slow when its your first day, its fun and all as long as everyone is helpful because if the work culture is not iterative, and blame game then it all falls apart oh yeah i never got to test my change before it hit the production as the tool was not built properly and local testing was a mess, so just like any other early tool "tested on production and lived happily ever after" jk