I Built a Fully Automated Affiliate Site in One Day. Here Is What I Learned.

I wanted to know if a fully automated affiliate site could be built solo in one day. By evening, the agent was publishing on its own.

The pieces are now cheap enough to assemble in a day, but what that produces still has to earn its place over months.

I did not set out to launch a business today. I wanted to know if it was possible to build a fully automated content and affiliate site on my own, in a single session, using the tools available in May of 2026. I had pieces of the system in my head from other projects this week. I had not stitched them together end to end, and I was not entirely sure I could.

By evening, the site was live at solobuild.io. The first article was written and published by an agent. Three affiliate programs had approved me within hours. The system runs on Monday, Wednesday, and Friday at 6 AM and writes a new article each time, without me touching the writing. None of that was the goal at 9 AM. The goal was to determine whether it was reachable.

This is the story of what actually happened. The specific build, the specific friction, what worked, and what an experiment like this actually proves.

Why this morning

The reason this morning was the right morning is that the tooling had crossed a threshold for me earlier in the week. I built my first real platform agent on Tuesday and wrote about it on Medium. The mental model for what a cloud agent is, where it lives, how it gets data, and what kind of decisions it can make was finally in place. That changed what felt possible to attempt in a day.

The niche I picked was AI tools for solopreneurs. That is a specific overlap with my own audience, and an honest one. I have spent the last year using these tools, paying for them, switching between them, and arguing with myself about which ones are worth keeping. A content site that helps other solo builders make the same calls is something I would read myself.

The shape of the bet was equally specific. A content site with an affiliate model. No email gates at launch. No paid acquisition. No analytics integrations beyond GA4. A scheduled agent that produces drafts I review before publishing. The whole point of scoping the day this tightly was that the speed had to come from constraint, not from rushing. I wanted the experiment to be honest, not stunt-shaped.

I did not know whether the affiliate applications would come through fast enough to matter. That was the real risk on the morning calendar.

The build

The domain registration on GoDaddy took twenty minutes. Solobuild.io was available, which I took as a small good omen. By the time the DNS records had propagated, the Lovable build was already underway in another tab. And before I get too clowned for using GoDaddy, I like the quick integration they have with Lovable.

I built the full site in Lovable. A tools directory, an articles section, an about page, and an admin panel. The design language is intentionally calm. Soft palette, generous spacing, no banner ads or popups, no urgency framing in the copy. The reason for that choice is positioning more than aesthetic preference. A site that helps solopreneurs choose tools cannot itself feel like it is selling them something.

I connected Supabase as the backend with three tables. Articles, tools, and subscribers. The articles table is the one that matters most because it contains both the published pieces and the drafts the agent will post. Each article has a status field that lets me move it from draft to published in one click from the admin panel. Each tool gets its own page with the affiliate link wired in.

The admin panel auths took longer than I expected, which is the part that always takes longer than expected on a Lovable build. Once the auth was working, the whole site felt operational by mid-afternoon.

Lovable and Supabase have become my default stack for a one-day site. The reason is that neither tool requires me to think about infrastructure. Both feel like they are getting out of my way, which is the highest compliment I can pay a tool.

The affiliate applications

The least technical hour of the day was the most strategically important. I applied to seven affiliate programs in one sitting. Lovable, Kit, Constant Contact, Writesonic, Surfer SEO, Notion, and Descript. The application step itself is mostly URL submission and a short “what is your site about” field.

The thing that helped most was that the site was already live when I was applying. Real domain, real pages, real content shape. A few of the affiliate teams check the site before approving, and a half-built landing page tends to bounce. A working site with empty article slots reads as a real attempt.

Three of the seven were approved the same day. The rest are pending and likely to come through over the next week or two. Same-day approvals matter more than they sound, because the agent could write its first article with real affiliate links rather than placeholders. That alone made the difference between a demonstration and a working thing.

One small observation. The affiliate companies that approved the fastest were the ones I have actually used myself. The slower programs are running on stricter review queues, which is useful information for sequencing the next batch of applications.

The webhook and the agent

The hardest part of the day, by a wide margin, was the agent. The webhook endpoint that lets an external agent post articles directly to the database. The system prompts that tell the agent how to research a topic, write an article, format the affiliate links, and post the draft.

The webhook endpoint is a secured HTTPS route on the site that accepts a structured payload from the agent. Title, subtitle, body, tool mentions, affiliate links, status. The endpoint validates the payload, inserts it into the articles table with a draft status, and returns a confirmation. The reason a webhook matters is that the agent is running on platform.claude.com, not on my Mac. It cannot access the database without a public endpoint to post to.

The content agent runs on Monday, Wednesday, and Friday at 6 AM. It researches a topic, drafts a full article, includes two or three affiliate links from the approved programs, and posts the draft through the webhook. The schedule is part of the system because consistency matters more than volume on a content site. Three pieces a week, indefinitely, with the same voice and structure, is a different shape than ten pieces in a burst followed by silence.

The agent posts drafts, not live articles. The human review step is the entire reason this system works. I read the draft, make small edits, and publish or trash it. If I publish, the affiliate links go live on a real page. If I trash it, the agent gets feedback through the system prompt the next time around.

The agent works alone three days a week, but every draft passes through a human before it goes live.

One thing worth naming. This whole section took roughly two hours. If I had tried the same stack a month ago, the webhook-plus-agent combination would have taken two days on its own. The reason it took two hours is that the article I published earlier in the week about building my first platform agent paid off in the most direct way possible. The mental model was already there, and the work just had to be done.

The first article

The first article the agent wrote and posted as a draft is titled "Best AI Writing Tools for Solopreneurs in 2026." The subtitle is “An honest look at the AI writing tools worth paying for in 2026, and the simpler stack most solo founders actually need.” It is now live on solobuild.io.

The draft that landed in the admin panel was usable on the first try. The structure was right. The tone was close to the site’s calm register. The affiliate links were placed where a real reader would actually want them, not stuffed into the opening paragraph or stacked at the bottom. The piece argued for a simpler tool stack than most content sites would push, which aligns with the site’s positioning.

I made one small edit. The opening paragraph was slightly stiffer than the rest of the piece, and I tightened it before publishing. Everything else stayed as the agent wrote it. The published version went live at the same time as the site.

The small, odd feeling of editing a piece I did not write, but that is still in my domain, is something I expect to get used to. The agent’s voice and my edits will converge over the next few weeks, as the system prompt sharpens based on what I cut and what I keep. The first article is the worst article. That is true for everyone, including the agent.

What this actually proves

The pieces are now cheap enough to assemble in a day. That is the first honest takeaway. A domain, a working frontend, a backend, a webhook, a scheduled cloud agent, and three approved affiliate programs are a stack that would have been a multi-week build two years ago and is now achievable on a calm Tuesday if the operator has done the adjacent work first.

The adjacent work matters. None of this would have happened today if I had not built a platform agent earlier in the week, if I had not been using Lovable and Supabase for months on other projects, if I had not been thinking about affiliate positioning all year. The day was fast because the runway behind it was already long.

The other honest takeaway is that an automated business in 2026 is still a content engine, not a money printer. The site has zero traffic right now. The articles have to be indexed. The pages have to rank for something. The affiliate clicks have to convert at a rate that justifies the time. None of that is solved by today. Today only proves that the experiment is real enough to run.

The agent will get better with each draft I review, because the system prompt will sharpen based on what I edit out. The voice will tighten. The link placement will improve. The topic selection will get more deliberate. Three months from now, the system will look almost the same and feel meaningfully better, which is how improvement usually works in systems like this.

What this does and does not say about the future of solo work is the part I am still working out. I am not going to claim that everyone can do this, because most people will not bother, and those who do will run into the same questions I am about to face. Indexing, ranking, distribution, conversion, affiliate program shifts, agent drift, all of it.

What I can say is that the experiment ran. The site is live. The agent is writing. The first article is published. That is the part that is no longer theoretical.

Thanks for reading, now go build something fun.

Eddie

P.S. The site is at https://solobuild.io. If you want to watch an automated content site, try to grow in public, that is where it lives. New articles publish Monday, Wednesday, and Friday at 6 AM.