I built a web directory in 15 hours and with $210 using the @johnrush method, and within a week, I have 38 pages indexed on Google and some organic SEO views trickling in. #buildinpublic
Here’s how I did it: 🧵
@brianoliveri No longer leading a tech team, unless you count agents, but used to. Father of two young boys and cranking to make everything possible dream of a furiously fast as possible. Nice to meet you!
You're not wrong. I do the same thing: each code repo keeps its own context in AGENTS.md and CLAUDE.md, and every agent on that project reads it there.
The separate repos are for stuff that doesn't belong to any one project. Mind-dump is my cross-project layer, where each project gets a status file so my bots can see what's happening everywhere, including projects that don't have a repo at all (family stuff, side projects, health). ai-archive is just raw backup of every Claude Code and Codex session so I own my history. It's huge and full of sensitive data, so I keep it private and away from my working repos.
So project context stays in the project. The other two are a dashboard and a backup.
Two years ago I built a web directory in 15 hours for $210. It's still my most bookmarked post.
Now AI agents run it. I check in and make the calls they can't. Google clicks were up 66% in October.
I'm including the entire playbook. Just tell your bot to read this thread. 🧵 #buildinpublic
🔟 The playbook
Tell your bot to read this, and it can set up the same thing for your own directory: https://t.co/aLwUiyb2X1
The full story, including the decisions the AI couldn't make: https://t.co/FaxS8sT6NB
@Mokhalil01@Rubio_tv I setup my machine to dump all the codex chat context into a repo, and keep another project repo updated with summaries of work done. GrokBot also accesses these repos, so context if any work done on one agent is available to another.