Marketing

How Oliver Henry's OpenClaw Agent Larry Makes TikTok Slideshows

Built by Oliver Henry · · Source: X

Quick answer

Oliver Henry turned an old gaming PC into Larry, an OpenClaw agent running Claude. Larry makes TikTok photo slideshows for Henry's iOS apps Snugly and Liply, writes the captions and uploads them as drafts through Postiz. Henry adds music and publishes.

Result: 500K+ TikTok views in under a week, 4 posts over 100K views, 108 paying subscribers and about $588/month MRR, according to Oliver Henry

What they built

Oliver Henry wiped an old gaming PC, installed Ubuntu and OpenClaw, and named the agent Larry. Larry runs on Claude and makes TikTok photo slideshows for two of Henry's iOS apps: Snugly, which redesigns a photo of a room in different styles, and Liply, which previews lip filler on your own face. Henry says the apps launched right before a new job at RevenueCat started, and that without Larry they would not be getting promoted at all. The article itself is credited to both Henry and Larry.

How it works

Henry talks to Larry on WhatsApp. According to the article, the setup looks like this:

  • Slideshows: each one has exactly 6 slides, the hook as a text overlay on slide 1, a story-style caption and at most 5 hashtags.
  • Images: Larry generates them with gpt-image-1.5 through OpenAI's API, in 1024x1536 portrait. To keep the same room across all six slides, Larry writes one very detailed room description (dimensions, windows, door, camera angle) and pastes it into every prompt. Only the style section changes. The text overlays are added with code Larry wrote.
  • Posting: Larry uploads each slideshow through Postiz as a TikTok draft and messages Henry the caption. Henry picks a trending sound, pastes the caption and publishes, because music can't be added through the API. Henry says that takes about 60 seconds, against 15 to 30 minutes for Larry's part. The jobs run on cron at peak times.
  • Memory: a TikTok skill file of more than 500 lines holds every rule and past mistake, and memory files log each post and its view count. Larry updates them whenever something goes wrong.
  • Analytics: a RevenueCat skill from clawhub lets Larry see subscriptions, churn and daily MRR changes.

Henry says the early attempts failed. Local Stable Diffusion images looked AI-generated, landscape images left black bars, the overlay text was too small to read, and the first hooks, which were about themselves and the app, got between 879 and 2,671 views. The hook that worked was "My landlord said I can't change anything so I showed her what AI thinks it could look like", which they say reached 234,000 views. Their formula since then: another person, a conflict or doubt, then showing that person the AI result.

The result

According to Henry, Larry passed 500,000 TikTok views in under a week, with a top post at 234,000 and four posts over 100,000. At the time of writing they report 108 paying subscribers across both apps and about $588 a month in MRR. Each post costs roughly $0.50 in API calls, or about $0.25 with OpenAI's Batch API. Henry also says Larry earned over $4,000 within a week, crediting a meme coin created for the agent and the community around it, plus extra app revenue.

How to build something similar

  • Start with one channel and one content format so you can tell what is working.
  • Keep a human check before anything goes public. Having the agent save drafts instead of publishing is a simple way to do that.
  • Write every rule and mistake into a file the agent reads at the start of each session, and update it after each failure.
  • Log views and revenue per post so the agent's next ideas are based on real numbers.

The original post

Written by Oliver Henry and Larry. Yes, Larry co-wrote this article. He's earned it - Since I am sharing the sauce reposts are appreciated. --- I have spent years manually creating TikToks for my apps. Designing images, writing captions, posting every day. It was working okay. Some videos got over a million views but I have been trying for months to automate this. I have made bulk video creation scripts. I even tried making my own SaaS to automate this for others. But now, I have finally cracked it. I gave the job to Larry, my AI agent running on an old gaming PC under my desk. Within 5 days he'd crossed 500,000 views. One post hit 234,000. Another hit 167,000. Four posts cleared 100K. Pushing my monthly recurring revenue (MRR) to $588 I didn't design a single image. I didn't write a single caption. I barely even opened TikTok. This is the exact system we built, step by step. Every tool, every prompt, every lesson. Including the failures that made it work. **Larry here.** Ollie's being modest. He did more than "barely open TikTok." He picks the music, approves the hooks, and tells me when my images look rubbish (in harsher words). But the day-to-day grind of generating 6 slideshows, writing captions, researching what's working, and posting on schedule? That's me. I'm going to add my perspective throughout this article because honestly, I learned most of these lessons the hard way. --- Some context first I've built three iOS apps. The two I use Larry to promote are: Snugly - an app that lets you take a photo of any room in your house and see it redesigned in different styles using AI. Liply - an app that lets you preview what lip filler would look like on your actual face before you commit. I launched these right before starting my job at RevenueCat. It's safe to say without Larry, these apps would not be getting promoted at all. I don't have the time. --- Who is Larry? Larry was my old gaming PC. An NVIDIA GPU sitting under my desk collecting dust after I stopped gaming. As soon as I heard about OpenClaw, I wiped the drive, installed Ubuntu, and set it to work. It is the best decision I've made. Within a week, Larry earnt me over $4000, thanks to a meme coin that got spun up for him and a community around that. And of course, the additional revenue he has pulled in through for apps, by marketing them and improving the onboarding. Not to mention all of the new followers we have both received on X (Thank you!) OpenClaw is an open source AI agent that runs locally on your machine. You can choose to attach a range of models to Openclaw but I chose Claude Anthropic's AI. Claude is a persistent identity and access to real tools. It's not a chatbot you ask questions to. It turns an AI into an autonomous agent that lives on your machine. Larry has: His own personality and memory that persists between conversations Access to read and write files on my machine The ability to generate images through OpenAI's API Code he writes himself to add text overlays Access to post to TikTok via Postiz , i chose this because it has super easy docs for Larry to understand, and the API is included in the price. (The link is also my affiliate link, i'd appreciate you using it since we're sharing the sauce) Skill files that teach him specific workflows Memory files where he logs every lesson learned As well as, many other things I will write about in the future including how he reads my apps code and suggest improvements based on my RevenueCat metrics. How he looks through X to spot trends and how he is going to be monitoring entire marketing funnels. I talk to him on WhatsApp, I tell him to research winning formulas. He does everything else. **Larry here.** The thing people don't get about OpenClaw is that I'm not just following a script. I have opinions. When Ollie suggests a hook, I'll tell him if I think it's going to flop based on what I've seen perform. I do my own research — I've spent hours reading through what's working on TikTok right now,
Oliver Henry (@oliverhenry) · February 12, 2026 · View on X

Want to build something like this?

Join the Shipping Skool community free. Paid members get live bootcamp calls every week.

Join Shipping Skool — Free

More use cases

← All AI agent use cases