You Shipped. Now What? A Marketing Workflow for Solo Founders

Your feature is deployed. The docs are updated. You close your editor, open X, and stare at the empty post box.

What do you say beyond “I shipped a thing”?

For founders building with AI coding agents, that handoff deserves its own workflow. The product already has useful raw material: documentation, examples, constraints and explanations of the problem it solves. Marketing starts with turning that material into something another person can use.

At APSquared, we're building promotion-agent, an MIT-licensed open source workflow for doing that work with your existing AI agent. It's designed for Codex, Claude Code and OpenCode, and it's still early access.

We introduced the project here. This post is about a small way to try it: one product, one useful draft, one next action.

Start with one problem your product solves

“Write marketing for my app” leaves a lot for an agent to guess.

A better starting point gives it a reader and a specific job:

Draft one X post for solo founders about the problem this product solves. Use documented capabilities. Flag anything you can't support.

Promotion-agent uses your product repository as a source and saves marketing work in a separate private workspace. You can review and revise in conversation, then return to the saved work later.

The useful test is simple: does the draft explain something your intended user cares about, and can you support every claim?

Turn a feature into an explanation

Consider a fictional app that helps small teams organize reference notes. Its documentation supports that capability. It doesn't support claims about hours saved, customer growth or being the best tool in its category.

A straightforward draft might say:

Give your small team's reference notes a home. Example Desk helps you organize them.

That's a starting point for editing. You might ask for a more specific audience, a clearer example or a different tone. If the agent needs a fact the docs don't contain, supply it or leave it out.

Try this follow-up:

Make it more useful to someone who hasn't heard of us. Explain the problem first, then show where the product fits. Don't add a testimonial or a performance claim.

Review the result as carefully as any other copy carrying your name.

Build an article around a reader's question

A feature announcement and a useful article can come from the same product knowledge, but they need different structures.

For the fictional notes app, an article could answer: “How should a small team organize reference material?” It could discuss ownership, naming conventions and when to archive old notes, with the product appearing where it helps explain the process.

Ask:

Propose an article answering one practical question our target users have. Check existing topics first. Separate advice we can support from examples or research we still need.

Promotion-agent includes a blog proposal workflow and history checks. Outside sources still need to be checked, and a draft still needs an editor. A product README can't establish every claim in an article.

The same principle applies when discussing other people's articles: contribute a useful observation, credit the source and connect your product only when it belongs in the conversation.

Keep the next action small

A folder of drafts can become another unfinished project.

Promotion-agent keeps tasks and history alongside the work. Ask “What's the status?” to see what's saved and what's pending. A prepared post remains a draft; a queued task still needs someone to act on it.

For a first attempt, keep the queue short:

1. Review one X draft.
2. Choose one article idea worth developing.
3. Decide what feedback would make the next draft better.

Publishing and outreach require separate authorization. The workflow runs during your conversations; there is no background marketing worker.

Try one draft on your own product

Follow the current setup instructions on GitHub. You'll need Git, Node.js 22.14 or newer, and a supported AI agent. Setup currently uses a source checkout (there's no npm package or one-command installer yet) and creates a separate private promotion workspace. The shortest path is to give your agent the repository's INSTALL.md and ask it to install promotion-agent for you; it asks a few questions and handles the rest.

Then ask:

Set up promotion for my project [product name] and draft one X post for review.

The software is free and MIT licensed. Your existing agent subscription or model usage may cost money; drafting needs no additional model API key, publishing account or database. Optional services can have their own costs. Private workspace storage also doesn't mean local-only AI processing: your chosen host and model provider handle the material the agent reads.

It's early access. Automated helper and build checks don't establish that every host's native invocation works, and an outside founder's first useful draft remains an acceptance goal.

If you try it, we'd like to hear something concrete: what you asked for, whether the draft was usable, what you had to correct and where setup was confusing. Share only public or sanitized examples through GitHub issues, or tell us on X at @apsquareddev.

That feedback gives us a useful next thing to build. Your first test can be as small as one post.