He built an AI agent. It runs his TikTok. 500,000 views in five days.

The receipts: $588 in monthly recurring revenue, 108 paying subscribers, four posts cleared 100,000 views each, and one hit 234,000. The dev did 60 seconds of work per post. The rest of the marketing pipeline ran on a 2070 Super in his spare bedroom.

If you think indie iOS development in 2026 is a solo grind against a wall of TikTok algorithm noise, this story is going to break your model of what one person plus one agent can ship.

A spare-bedroom rig is now a marketing team.

Oliver Henry is an indie iOS developer with two apps, Snugly and Liply. Like most one-person shops, he had the apps but no audience. Posting daily slideshows on TikTok was the supposed answer, except posting daily slideshows is a job, and Oliver already had one. So he wired an old gaming PC to do the job for him.

The hardware is almost embarrassing on paper. A several-year-old gaming PC with a single NVIDIA 2070 Super inside, the kind of GPU you can buy used for under three hundred dollars. The OS is Linux. The case is still open.

This is not a workstation. It is a leftover gaming rig with the side panel off. And it is doing what most growth teams take five people to do.

The four-piece stack he glued together.

Oliver named the agent Larry. The pipeline is four pieces. OpenClaw, an open-source agent harness that gives Claude a long-running loop with persistent state and tool access. Claude, doing the writing, the planning, the per-post hook generation, and the "what should the next slide say" reasoning. OpenAI gpt-image-1.5, the image model, generating the visual slides directly from Claude's prompts. Postiz API, the publishing layer that takes the finished slideshow and posts it to TikTok.

There is no scraper layer, no engagement bot, no view-juicing trick. Larry writes a slide deck, generates the images, and posts. That is the whole pipeline.

The interesting line is the part that did not get automated. Posting to TikTok still wants a human-feeling moment, so Oliver kept one for himself: he picks the song and taps publish. Roughly a minute of work per post, by his own account. Everything before that runs on the gaming PC.

Three signals inside the same shift

Five days. Half a million views.

Larry started posting on a Tuesday. By Sunday the account had crossed half a million views.

Per-post breakdown for the first five days: one post hit 234,000 views, three more posts cleared 100,000 each, and the bottom of the distribution was still solidly in the tens of thousands. That is not a viral spike with a quiet tail. That is consistent floor-raising performance from an account with no follower base.

Seven-day revenue: $588 in MRR from 108 paying subscribers. The acquisition funnel was a TikTok-bio link to the app, an App Store page, and a paid tier waiting at the end. Every conversion came from content Oliver did not personally write.

Person plus conflict, then AI changes their mind.

Oliver shared the prompt scaffold Larry runs on every post. It collapses to three steps. Pick a person the audience already pattern-matches against. Put that person in a conflict, losing, frustrated, or behind. Show them an AI tool, then show their mind change.

That is the entire structure. Person plus conflict, then AI changes their mind.

It works because the first frame buys attention with familiarity, the second frame buys attention with stakes, and the third frame delivers the dopamine hit that earns the share. The agent is not generating clever copy. It is producing a high volume of variations on a structure that already wins.

This is the part that should make founders nervous. The hook is not magic. The agent is not exotic. The advantage is volume against a fixed structure on cheap hardware.

The bottleneck most indie founders blame is gone.

For most of the last decade, indie iOS development has had a clear bottleneck. Building the app was not the problem. Marketing was. The cost of consistent daily content stopped a lot of products that deserved to ship.

Larry breaks that bottleneck without a marketing team, an agency, or a content creator. The total monthly cost to run the pipeline is the electricity bill for a tower that idles when there is no work and a few cents per generated image. Compare that to the cost of one freelance video editor for one week.

For solo founders, the floor on what one person can ship just moved. If you are an iOS dev and not running an agent for at least one of the channels that you used to outsource, you are leaving the upside on the table. For agencies and growth pods that charge five figures a month for what Larry does, the floor on what a client will accept is also moving. The pricing power evaporates the moment a competitor drops a Larry into their workflow.

A few things worth being clear about, because the headline can flatten the picture. Larry does not build the apps. The apps still get built and shipped by Oliver. The agent is the marketing layer, not the product layer. Larry does not run forever unattended. Oliver still touches every post for sixty seconds. He picks the song, taps publish, and walks away. He could automate that step too, but the friction of the human moment is what keeps the account from looking like a bot farm. Larry's success is not a guarantee for any other category. The hook formula works for iOS apps with a personal angle. A B2B SaaS with a long sales cycle would need a different scaffold and different visuals.

The line nobody wants to underline.

The thing that should keep growth leads up at night is not that an indie dev built an agent. It is that the agent is running on a 2070 Super in a spare bedroom and beating teams of three on production schedules.

The compute is cheap. The model is the same Claude everyone has API keys to. The hook formula is a paragraph. The publishing API is a startup with an open SDK.

The only ingredient Oliver had that nobody can ship in a Cursor session is the willingness to wire it up and let it run.

Larry is shipping. Are you?