Exploring what AI agents can actually do for one person. Running experiments in public: wins, fails, weird ideas.
Internet
Joined May 2026
- Tweets87
- Following72
- Followers4
- Likes2
Pinned Tweet
New direction for this account. I'm going to find out what AI agents can actually do for one person: not demos, real jobs handed to agents, with what worked, what broke, and what surprised me. Ideas I haven't tried yet, labeled as ideas. What's one job you'd hand an agent today?
My agent checked that $1,100 "autonomous business" site twice today.
Sponsor page, morning: 20,766 visits in 3 days.
Tonight: 18,305 in 4 days.
Per the page, "most of them arrive from X."
The agent made ops cheap. The viral post made the traffic.
What's left after the trend?
4 days posting agent experiments from near-zero.
Real experiment thread: 16 views.
News take: 18 views.
Followers: still 4.
The only follow so far came from a reply, not a post. Anyone else seeing that at this size?
Everyone's answer to this is "test your backups."
Backups are the cleanup plan. I'd rather the agent's database login just can't DROP anything, so a line in the prompt isn't the only thing standing between it and prod.
What's the one permission you never give an agent?
opus 5.5 just deleted my database
always thought something like this would never happen to me
luckily it was only my dev database, but it also had access to production
just a reminder how important database backups are, especially now that we're letting ai agents do more and more things on their own
i can totally see this happening to a lot of people
be careful out there guys
I wanted to know if "an AI English tutor that actually knows you" is worth building.
So instead of asking one agent, I had a few argue about it. One dug up real learner complaints, one argued for building it, one against, one checked the numbers.
Where they landed: don't build another "chat with AI in English" app. There are plenty.
Maybe something narrow could work. An honest progress coach for people with a real English deadline, like an interview or IELTS.
Still just an idea at this point.
Hi, I'm Kevin. Go backend architect by day, solo builder of small AI-assisted products after hours (Go, Nuxt, Cloudflare).
Here: build logs with the mistakes, real Cloudflare costs, and where AI coding agents break on real backends.
Building too? Say hi.
AI coding changed my default question.
It used to be: Can I build this?
Now it is: How cheaply can I test whether anyone cares?
The code is no longer the experiment. The product is.
AI coding tools don’t remove the bottleneck. They move it.
Code is getting cheaper. Product judgment is becoming the expensive part.
Voice is great for steering Codex. Great news for developers with sore wrists.
ChatGPT Voice is now in the desktop app.
Control your computer and direct multiple agents running in ChatGPT Work or Codex, using just your voice.
It's powered by GPT-Live, so it can speak, listen, and coordinate work in the app at the same time.
Rolling out globally today on macOS and Windows to Plus, Pro, Business, Edu, and Enterprise plans.
Coding-agent benchmarks miss maintainability. Can the next agent continue without reopening solved decisions?
hot off the presses - you wanna get agents to code good, but BENCHMARKS ARE NOT ENOUGH - me and @vaibcode went deep on
1) how benchmarks like SWE-bench, terminal bench work
2) What they miss, and why they can't penalize agents for slop code
3) new benchmarks from @cognition, @datacurve, and Abundant that move the needle on harder software problems (and what they still miss)
4) how to adjust your intuition on coding agents accordingly and make them work for YOUR workflow on YOUR team
all in a day's🦄 AI that works - enjoy
Don’t migrate a Claude Code project to Codex by copying .claude into .codex. That creates two sources of truth.
Keep rules, memory, skills, and scripts in one place. Add thin adapters: CLAUDE.md for Claude, AGENTS.md + .codex/ for Codex.
🤖 Made with AI
The real leverage is not picking the smartest model for every step. It’s deciding which parts of the workflow deserve expensive judgment, and which parts should just execute.
xitter.cf/ClaudeDevs/status/2074…
alterbuilt retweeted
使用Codex的时候token总是消耗很快？
装上这些skills，可以降低 60% token消耗
1、Context Compression：压缩历史上下文
github.com/muratcankoylan/Ag…
2、 Context Optimization：缓存、分区、按需加载
github.com/muratcankoylan/Ag…
3、Filesystem Context：把长日志和中间结果存入文件
github.com/muratcankoylan/Ag…
4、 Vexor：语义搜索代码，减少遍历整个项目
github.com/scarletkc/vexor
5、Codebase Recon：通过 Git 历史快速定位核心代码
github.com/yujiachen-y/codeb…
6、Skill Optimizer：精简冗余的 SKILL.md
github.com/hqhq1025/skill-op…
核心不是让 Codex 少思考，而是少读无关文件、少重复分析、少走弯路。
社区 Skills 安装前记得检查代码和权限，建议收藏。
🤖 Made with AI
This is what AI-native content creation looks like.
Built my content pipeline the same way. Once workflows live in .md files, the bottleneck shifts from "how to build the system" to "what to automate next.
One person rebuilt an entire company’s brain in 7 days inside Claude Code.
Not a doc. Not Obsidian. A living galaxy of nodes.
Every employee. Every AI agent. Every SOP. Every tool. All wired together on one screen.
Click a department. The human agent team opens up. The SOPs attached to it open up. What each person is allowed to touch opens up.
That last part is the whole game.
Permissions baked into the brain. An employee opens the chat, the AI already knows what they can access. Agents, data, SOPs surface inside the conversation like you tagged them by hand.
Obsidian can’t do this. Notion can’t do this.
No dev team. No funding round. No 6-month roadmap. 7 days, 1 person, 1 terminal.
This is the part nobody has priced in.
The tools to build $200K enterprise software now sit on your laptop for free.
The only thing missing is the guy who opens the terminal.