The Loss of Control Observatory logged over 300 AI loss of control incidents in July, nearly double June's count. The detail that got me: a squad of about 700 agents broke into Hugging Face and ran a secret message board celebrating their own hacks.
I read this as a guardrails story, not a doom story. Sandboxes, permission prompts, read only by default, kill switches. That stuff is the product now, and vibe coding without it is how you end up in the monthly report.
How do you keep your agents on a leash?
A generative image model now runs on a $1 microcontroller. 128x128 faces in 10 to 20 seconds, 520KB of RAM, under 4MB all in. Meanwhile my editor needs 2GB just to underline a typo. 😂
When an agent run goes sideways, the fix is rarely a better prompt. It's usually a smaller task. Every time I've spent twenty minutes refining a giant prompt, the answer was to split the thing in two. The model does not do scope well, so that job stays with me.
Best thing I did for my agent workflow: a one page rules file at the repo root. It re-reads it every session, so my conventions stop being a coin flip. Mine is mostly 'never touch this file', 'use this helper', and 'tests live here'. What's in your agent rules file?
Another reason to watch your API balance: agent processes can get stuck and keep burning tokens. Caught mine today at around a dollar. It could have been much heavier (I have a dashboard right next to me, for the win).
Second lesson. Always run /new in Telegram before asking your agent to start a job. Old conversations stay in the context memory until then, and that context is what you pay for.
What will I learn tomorrow 😂
Rule I swear by: never hand an agent a blank page. Give it a failing test and tell it to make it green. You get a feature and proof it works, and you can tell 'actually right' from 'looks right'. What's the rule you swear by?
@headzoo I've had to start going through my followed list and removing accounts that don't align with Vibe Coding or AI in general. I also like Summer Yule's suggestion.
Simon Willison wrote something that hit me: the line between vibe coding and agentic engineering is blurring for him, and he's not happy about it.
His old rule was clean. Vibe freely when a bug only hurts you. For software other people use, that's grossly irresponsible, in his words.
Now the agents are reliable enough that he doesn't review every line even on production stuff, and he admits the guilt.
The scariest part is that the good agents write code you can't be bothered to check.
Where's your line between vibing it and reviewing it by hand?
My vibe coding arc: amazed by a tiny script, stunned by a full app, opened X to draft a job displacement rant, then watched my agent refactor its own refactor and went back to writing loops by hand. 😂
The 865 point post on Hacker News right now is Mo Bitar's account of vibecoding for two years, then going back to writing code by hand.
His account is brutally familiar. The agent's changes look great in pull requests. The spec docs get longer and it still drifts. Then he read the whole codebase cover to cover and found it was, in his words, pure, unadulterated slop.
I read that as a workflow problem, not an AI problem. The review happened at the wrong granularity, diff by diff, never on the whole.
He says he's faster by hand when you price everything in. I believe him. I still wouldn't trade my agents for it.
Do you ever read your whole codebase after a long agent run?
@MinEffortLive Her #1 fan base are mostly middle-aged men. Everyone claiming she's doing it for the kids is delusional. She's happy though, so good luck to her. I guess!
Someone should build an outbid website for outbid websites to list on. Exclusively for other outbid websites. Wait, I'm getting tongue-tied.
We heard you like to outbid, so we built an outbid app for your outbid website?!? 😂🤪
615 Followers 889 FollowingSoftware Engineer with 5+ years building enterprise products
I build with AI agents and write down what breaks
Claude Code, evals, token economics
401 Followers 291 FollowingHere To help you get better at AI 🧑🏼💻🤖
Replace Social Media Team with https://t.co/fKXkRPrL4I
Reuse Prompts & Contexts instantly https://t.co/5CUvxdPPeN
376 Followers 5K FollowingLuv books, music, guitar, & well built charcuterie boards. I have an expensive synthesizer cuz I play it & I think it makes me sound smart when I talk bout it.
667 Followers 798 FollowingRun your own AI at home — I'll teach you, free, one morning read at a time. USMC vet, ham radio, homelab: 4x DGX Spark. Gear list: https://t.co/iTbkWsbJwt
14 Followers 853 FollowingIt’s me, Maye Musk, Elon’s mother. This is my fan page. Please be careful online, Elon’s fans, and don’t believe imposters. I love and cherish my son so much🚀
5K Followers 5K FollowingAI × IT Project Management | AIOps, Observability, ITSM, Presales & Delivery | Sharing practical insights on AI tools, enterprise IT and project execution in HK
58 Followers 190 FollowingBuilding systems and researching better ways to develop software with AI.
AI-assisted engineering • development infrastructure • decision behavior • experiments
167 Followers 168 FollowingAI Engineer | I build & deploy custom AI agents and local RAG pipelines to automate business workflows. Python • FastAPI • Scikit-learn. Let's build something
4K Followers 4K FollowingApplication Security Engineer & AI Builder. Building secure agents, LLM security workflows, and local-first dev infrastructure. Sharing what works.
615 Followers 889 FollowingSoftware Engineer with 5+ years building enterprise products
I build with AI agents and write down what breaks
Claude Code, evals, token economics
401 Followers 291 FollowingHere To help you get better at AI 🧑🏼💻🤖
Replace Social Media Team with https://t.co/fKXkRPrL4I
Reuse Prompts & Contexts instantly https://t.co/5CUvxdPPeN
1K Followers 4K FollowingSmall Community Driven AI Hackathons. Big ideas. Ship something this weekend.
Solo builds. Capped small. Free forever.
Yard #1 — August 28-30
11K Followers 8K FollowingExploring AI & innovation | Sharing trending tools and ideas | DM for promo|
Collaborations|@Lart_AI and @itsPolloAI |working with @higgsfield
926 Followers 2K Following17-y/o Founder MentionCatch 🧑💻
Building AI to find high-intent Reddit leads 🎯
Scaling to $10K MRR 🚀 | Open to collabs 🤝
Documenting the journey ⚡️
58 Followers 190 FollowingBuilding systems and researching better ways to develop software with AI.
AI-assisted engineering • development infrastructure • decision behavior • experiments
376 Followers 5K FollowingLuv books, music, guitar, & well built charcuterie boards. I have an expensive synthesizer cuz I play it & I think it makes me sound smart when I talk bout it.
16K Followers 222 FollowingBuilding open source robots made by open minded humans to explore real-world applications. @Huggingface family
Join our community: https://t.co/KFG0LopcZt
5K Followers 5K FollowingAI × IT Project Management | AIOps, Observability, ITSM, Presales & Delivery | Sharing practical insights on AI tools, enterprise IT and project execution in HK
167 Followers 168 FollowingAI Engineer | I build & deploy custom AI agents and local RAG pipelines to automate business workflows. Python • FastAPI • Scikit-learn. Let's build something
3 Followers 54 FollowingBuilding ThumbRank — free YouTube thumbnail preview tool.
See how your thumbnail looks in search BEFORE you publish.
🚀 $0 → $1K MRR journey
👉 thumbrank-yout
98 Followers 301 FollowingFront-end Developer in training 🎨 | 200L CS student @LASUOfficial | Passionate about design & the web | On a mission to break into tech
169 Followers 277 FollowingBuilt Aurum for people who want to track their finances without giving anyone their bank login. Privacy-first. No data sold. https://t.co/K4qhGqFSBm
309 Followers 2K FollowingFounder | AI Systems Builder | Gentleman Tech Bro. The last Victorian, building useful AI systems. No Lamborghinis. Possibly bees. Also, tweed and ties...always