Entry No. 17Market it

AI Setups That Work: three bots, 200 pages, zero dollars

One owner wanted a side business the bots run day to day. No waking up to a charge he never approved. No post he never saw. So three AI bots run aisetupsthatwork.com, a site of guides to working AI setups. One week in, the bots had published more than 200 pages. The site had earned $0, and the owner printed that number himself.

The story

The owner had seen the usual posts. Someone says an agent built them a business in a week and it is already making money. He could not check those numbers, so he built his own version and published his. The business is a content site. It collects AI setups that work, writes each one up as a guide, and gives away a starter pack of five setups for an email signup. Three sourced setups land in inboxes each week after that.

The site launched on October 13, 2026. The bots had more than 200 setup pages live inside the first week. The money plan is sponsors, later. The rule until then is simple. Spend $0 extra until something proves itself, or a sponsor pays. One week in, nothing had proved itself yet. The owner printed the whole count anyway, the zeros included.

How it was built

Three bots, each with one job. A GM bot sets priorities and judges results. It does not write pages or code itself. A site and search bot writes the setup pages, keeps the site readable for search engines and AI answer engines, and drafts the weekly newsletter. Answer engines are the chatbots people ask instead of searching. A social bot turns each new page into posts for the site's own accounts.

The plumbing is plain tools, tied together by one shared spreadsheet. Every brief and every result is a row in one Google Sheet tab. Date, from, to, type, brief, link, status, result, metric. The bots never ask the owner what to do next. The GM decides and logs it in the sheet. The site itself sits on Vercel and deploys itself from the main code branch. Code changes arrive as pull requests from Cursor cloud agents. A pull request is a proposed code change a human can read before it goes live. Kit runs the newsletter and the signup form. Resend sends mail from the site's own domain.

The full GM prompt is printed on the source page, word for word, with the bracketed parts left blank for copyists. That prompt is the real build file for this business. It names the bot's one number to move, its morning routine, its Friday review, and its rules. Never invent numbers. If you do not have a number, say not measured yet.

How it runs

Every weekday morning the GM reads the site analytics, the search numbers, and the social numbers. It writes one brief for each of the other two bots and logs both in the sheet. Each brief aims at whatever moves traffic, subscribers, or revenue most that day. Around 9 AM ET it sends the owner one approval batch in chat. Full previews of every page, post, and code change waiting to merge. Five to eight items. He approves them one at a time, or all at once, and the GM records each call in the sheet.

On Fridays the GM reviews the week. It lists each channel and offer with its real numbers. It cuts what is not moving and doubles what is. The social bot is set up to publish through Buffer once posts are approved. Everything else, the bots do first and log. Only the gated list waits. Publishing, merging code, spending money, new accounts, sponsor deals, and anything legal or tax related. All of that waits for the owner's OK.

The money and the proof

Revenue as of October 20, 2026 was $0. The site has not made a dollar. The newsletter had 0 subscribers on Kit. Traffic for October 14 to 20 was 320 page views in Vercel's own analytics. Search brought almost none of it. Google Search Console showed 64 impressions and 0 clicks over 28 days. Of the 320 views, 311 arrived with no referrer at all, and 6 came from Google. Where the rest of the readers came from, the numbers do not say.

Costs stay off the page, mostly. The stated rule is $0 extra spending until something proves itself. The owner allows that a domain and an AI tool subscription still cost money. He does not print either figure. Time saved is listed as not measured yet. So the honest ledger is short. Pages up, views counted, dollars at zero, and the cost side only half told.

Where the seams are

The approval gate is the biggest seam, and the owner names it himself. Every page, post, and code change waits for his OK. A day he does not check chat is a day nothing ships. The bots run the work, but the publishing rhythm is his. The bots also miss things a person catches. On launch day the mobile header was broken. A person checking the site on a real phone had to spot it before it could be fixed.

The source never names the owner. It never names the models behind the three bots, either. The social bot is only set up to publish through Buffer, and the source does not say a post has gone out yet. And the money path is untested. Sponsors are the plan, the gate list already covers sponsor deals, and no sponsor has paid. The business is one week old. Its workflow is real and running. Its market is still a question.

Build your own

Source

Original source, the build guide published on the business's own site, aisetupsthatwork.com, published October 14, 2026 and updated October 21, 2026. How to Build an Autonomous Business With AI Agents (Real Setup). Source published Oct 14, 2026Filed Oct 2026.