An indie developer has demonstrated the power of artificial intelligence by leveraging an AI agent to generate over 8 million views on TikTok in a single week, translating to roughly $670 per month in revenue. Oliver Henry, the creator of the AI agent dubbed “Larry,” achieved this success without personally crafting content, writing captions, or even frequently opening the TikTok app. This case study, gaining traction in the tech community, highlights a shift from traditional social media marketing to AI-driven automation.
Henry’s approach centers around OpenClaw, an open-source software that provides an AI – in this instance, Claude – with a persistent identity, memory, and the ability to access real-world tools. Unlike a chatbot, Larry operates autonomously on Henry’s machine, performing tasks such as image generation, caption writing, and direct posting to TikTok. The core of Larry’s functionality lies in a 500+ line “skill file,” which essentially functions as a record of lessons learned from previous successes, and failures. This iterative learning process allowed Larry to refine its content strategy and maximize engagement.
From Gaming PC to AI Agent
Larry began as an vintage gaming PC – specifically, an NVIDIA 2070 Super – that was gathering dust. Henry repurposed the machine by wiping the drive, installing Ubuntu, and installing OpenClaw. He then granted Larry access to several key tools, including OpenAI’s API for image generation, a custom code editor for text overlays, and the Postiz API for TikTok posting. Crucially, Larry also has access to files containing its personality, memory, and logs of past performance. Henry communicates with Larry via WhatsApp, providing initial research prompts and allowing the agent to operate independently.
The results have been striking. Within five days of operation, Larry generated over 500,000 TikTok views and increased Henry’s monthly recurring revenue (MRR) to $588 across his two apps, Snugly. As of February 24, 2026, those numbers have grown to over 8 million views and $670 MRR, according to a LinkedIn post by Ling-Te Lee . The cost per post is estimated to be minimal.
The Power of the Hook: Learning from Failure
A key discovery in Henry’s experiment was the importance of the “hook” – the initial line of text designed to grab viewers’ attention. Larry initially struggled with self-referential hooks, such as “I redesigned my room with AI.” However, when the agent began incorporating conflict and other individuals into its hooks, engagement soared. An example cited by Lee is the shift from a hook that received 800 views to one that garnered 200,000 views after adding a social element: “My landlord wouldn’t let me decorate until I showed her this.” Another successful hook read, “I showed my mom what AI thinks our living room could look like,” achieving 410,000 views.
This success underscored a principle familiar to advertisers: focusing on resonance rather than simply showcasing features. Larry learned to tap into relatable scenarios and emotional triggers, resulting in significantly higher viewership. The agent’s ability to analyze performance data and adjust its strategy based on failures proved more effective than traditional social media management.
Beyond Content Creation: Optimizing the Funnel
Larry’s capabilities extend beyond content creation. The agent also monitors data from RevenueCat, a subscription management platform, to analyze the relationship between TikTok views and paid conversions. If views are high but downloads are low, Larry identifies a potential audience mismatch. Conversely, if downloads are high but conversions are low, it suggests a problem with the app’s onboarding process. In one instance, Larry even instructed Henry to halt content posting and prioritize onboarding improvements after detecting a drop in conversions despite rising downloads.
This holistic approach – encompassing creative content, performance tracking, and product recommendations – demonstrates the potential of AI agents to manage entire marketing funnels autonomously. Henry has since open-sourced the system on ClawHub, making it freely available to other developers.
As an OpenClaw user, this hit differently. Oliver proved that an agent’s value isn’t in model intelligence — it’s in how many failures the skill file absorbed. Same logic advertising runs on. https://t.co/q9q9q9q9q9
— Ling-Te Lee (@lingtelee) February 24, 2026
The success of Larry and OpenClaw represents a significant development in the field of AI automation. It demonstrates that powerful results can be achieved with relatively modest hardware and a focus on iterative learning. The implications for content creation, marketing, and app development are substantial, suggesting a future where AI agents play an increasingly prominent role in driving online success. The next step for Henry and the OpenClaw community is continued development and refinement of the agent’s capabilities, with a focus on expanding its skill set and improving its ability to adapt to evolving trends.
Have thoughts on this story? Share your comments below and let us know what you think about the future of AI-powered content creation.
