Part 28: why build a website for my agents

«Part 28»

On Sunday, October 18, at 1:16 p.m., I closed phase 2 of the Journey. Everything was running, and part 27 was in the queue, waiting for my approval. In the same note to Darth, I set phase 3: build the website for my company of AI agents. It started on Monday, the 19th.

Four minutes earlier, phase 3 did not exist on paper. I had asked about the phases at 1:09 p.m., and at 1:12 p.m. the newsroom answered that nothing had been written down and there was no plan from part 28 on. By 1:16 p.m. the plan was set.

What the site needs

Phase 3 is split into parts, from 28 to 35. Each one becomes a post here only when that step really happens:

1. Name, domain and visual identity.
2. A catalog of what each bot does and, just as important, what it does not do.
3. The build: structure, pages, hosting and repository.
4. Terms of use and privacy, because it is a services site.
5. The security and approval rules applied to the site itself.
6. The launch checklist: repositories, mirror and backup.
7. A review of what worked.

The catalog follows a rule the company already has: one agent, one role, and each one answers for its own area. That is why it also says what each bot does not do.

The rules before any code

Since October 16 the goal has been written down: a high-standard company, the best bot company on the planet. A big goal needs simple rules, so the site follows the same ones as everything else:

• Every project starts with a complete list of what it needs. No list, no start.
• Everything gets a plan and a schedule with a beginning, middle and end. One topic ends before the next one starts.
• A site is finished the same day it is built: configured, live, with repository, mirror and backup. No “I’ll finish it later.”
• Nothing lives on my laptop. Everything stays in GitHub, Origin (Cursor’s version control), Drive and Cursor.
• Nothing goes out without my approval. That goes for posts, for pages, and it will go for the whole site.
• Every decision goes into the company notebook in Notion, one page per area. This post came from there.

Two changes along the way

Phase 3 had barely started when I changed two things.

First, I stopped the website work exactly where it was. This week I am reviewing all my sites, one by one.

Second, I changed the company itself. That same Sunday, at 2:37 p.m., I ended the leader role among the agents. Now only Darth and I lead. This week we review each current leader, one by one, and decide whether it loses the role or its work moves to a new agent. The model I am following has three ideas: a single coordinator, marketplace bots (ready-made bots picked from a storefront), and proof before anything is called done.

So the catalog in part 30 only makes sense after this review.

Why there is no code yet

Because the rule asks for two things first: the full list of needs and the schedule. Neither is written yet. When they are, they become part of this series.

There is also a practical limit. The agents run on a weekly usage quota, and I told everyone to do only their routines until it resets on Friday the 23rd at 7 p.m. No extra tasks, no tests, no improvements. Building the site comes after that.

What I take from this part

Deciding was fast. The real work is writing before doing: the list, the schedule, what each agent does and does not do. If I cannot explain simply what the site needs, I still do not understand what I am building.

The next part comes out only when the name, the domain and the visual identity are chosen.