Building a business a week with AI, in public

I've spent about 15 years in digital marketing and analytics, and over five years building with AI in the loop; the longer version is on the about page. The lesson that whole stretch taught me fits in one sentence: AI doesn't make businesses succeed, it collapses the cost of trying. Things that used to take a team and six months, one person can now design, build, and launch in a week. Once that's true, the rational move is to run a lot of attempts, and the interesting move is to do it where everyone can watch. That's SideRoad.

Why a week? Why public?

A week is a forcing function. It's long enough to ship something real, a product, a site, a working checkout, and short enough that you can't gold-plate, over-research, or hide in planning. Every corner the deadline forces you to cut teaches you which corners mattered.

Public is the accountability layer and the honesty layer at once. Anyone can claim a method works; publishing every build's revenue, the flops at their full flop size, removes the option of curating the story. It also happens to be the best marketing I've found, because documented reality is rarer than polish. The numbers live on the revenue tracker, updated for every build, wins and zeroes alike.

What does the week actually look like?

• Monday, decide. Pick the idea, define done, and write the founding brief with Claude interrogating the vague parts. The build's jotthread.md gets created here, and every decision from this day forward lands in it.
• Tuesday to Thursday, build. Product and site in Claude Code, copy and positioning in Claude, assets wherever they're fastest. The habits that make this pace survivable are documented in the Claude Code tips post.
• Friday, launch. Checkout live, site deployed, announcement out to the newsletter and wherever the build's audience actually is.
• The write-up. What shipped, what it made, what I'd do differently, published with the numbers. Then the next week starts.

AI's role across all of it is the same role: it's the team. Researcher on Monday, engineers midweek, copywriter on Friday, and the reason the week is enough.

What do the numbers honestly look like?

Uneven, and that's the design. Some builds earn from week one, some trickle, some flop completely, and all of it is on the tracker rather than summarized flatteringly here. The model is a portfolio: each build costs a week, stays alive after launch, and keeps its small chance of compounding, so the portfolio's value grows even while any single week looks unremarkable. The two assets that only go up are the audience that watches and the system itself, because every build makes the next one faster. The honest anti-hype line: a week can manufacture a launch, never traction. Traction arrives on its own schedule or doesn't, and pretending otherwise is how build-in-public curdles into performance art.

Where does JotThread fit this story?

It's the method eating its own output. Around fifteen weeks in, the operational problem of the whole system was obvious: living in Claude conversations all day means decisions, follow-ups, and settled numbers pile up mid-chat faster than any notes habit catches them, and losing Monday's decision by Thursday costs real time at this pace. So one week's build was the fix: a capture command, one Markdown file per build, a board to see it all. I've run every build since on it, and it's for sale the way everything here is, small, priced once, and yours to keep.

That's the part of build-in-public that compounds most reliably: your own workflow keeps handing you the next product, because the problems you hit at week fifteen are the problems everyone else hits eventually.

How would you start this week?

Smaller than feels impressive. One build, one week, one honest write-up with the real number in it, even if the number is zero, especially if the number is zero. Constrain the scope until a week is genuinely enough, let AI be the team, and capture your decisions as you go, because the write-up is where the value concentrates and it's unwritable from memory. If you want the running version of all this, the build archive and the weekly email are where it happens.

Common questions