The Larry Loop: How One OpenClaw Agent Got 2 Million TikTok Views | OpenClaw DC

Oliver Henry’s OpenClaw agent “Larry” generated 500,000 TikTok views in 5 days by autonomously creating slideshow content, reading analytics, and iterating on what works. Ernesto Lopez scaled the same strategy to $70K MRR. Here is how the Larry Loop works and whether you can replicate it.

If you are new to OpenClaw, read what OpenClaw is and how to install it first. If you want a broader picture of how people are earning with it, see our guide on how to make money with OpenClaw.

How the Larry Loop Works

The Larry Loop is a five-step cycle that runs continuously. Each step feeds into the next, which is why it is called a loop and not a pipeline.

Step 1: Content creation. Larry generates slideshow images using AI image generation, writes hook titles designed to stop the scroll, and drafts CTA text for each slide. The agent pulls from trending topics and formats that are already working on TikTok.

Step 2: Upload as draft. Larry uploads the completed slideshow to TikTok as a draft. It does not publish. This is a critical design choice. The content sits in the drafts folder waiting for human approval.

Step 3: Human review (60 seconds). Oliver opens TikTok, reviews the drafts, selects background music, and clicks publish. This is the only manual step. It takes roughly 60 seconds per day.

Step 4: Read analytics. After the content has been live for a set period, Larry pulls the performance data. Views, watch time, shares, comments, profile visits. It stores this data and looks for patterns.

Step 5: Iterate and repeat. Larry compares what worked against what did not. Hooks that drove higher watch time get weighted more heavily. CTAs that generated profile visits get reused. Formats that flopped get dropped. Then the loop starts again at Step 1 with updated knowledge.

The power of the Larry Loop is not in any single step. It is in the feedback mechanism. Most content creators guess what works. Larry measures, adjusts, and compounds its knowledge over every cycle.

What Larry Actually Does Under the Hood

Larry is not a simple script that posts on a schedule. It is an OpenClaw agent with several distinct capabilities working together.

Image generation. Larry generates slideshow images using AI models accessible through OpenClaw. Each image is formatted to TikTok’s vertical aspect ratio with text overlays positioned for maximum readability on mobile screens.

Copywriting. The agent writes hooks, slide titles, and CTAs. It does not use generic templates. It analyzes what language patterns are driving engagement in its analytics history and generates new variations. Early in the loop, this is mostly pattern-matching from trending content. After a few cycles, it starts producing hooks informed by its own performance data.

Analytics parsing. Larry connects to TikTok’s analytics and pulls structured data. It tracks which hooks correlate with higher watch-through rates, which CTAs drive the most profile visits, and which posting times get the best initial traction.

Draft management. The agent uploads finished slideshows as drafts rather than publishing directly. This keeps a human in the approval chain and reduces the risk of posting content that could get the account flagged.

Oliver described the system as “a content intern that works 24 hours a day, learns from every post, and never asks for a raise.” The difference from a real intern is that Larry does not get bored, does not have creative ego, and does not forget what worked last Tuesday.

The Human Part Is Small but Essential

Oliver spends roughly 60 seconds per day on Larry’s output. He opens TikTok, scrolls through the drafts Larry uploaded, picks background music that fits the content, and taps publish.

This is not laziness. It is a deliberate design decision. TikTok’s algorithms and policies are more favorable to content that has clear human involvement. Auto-posting with zero human interaction increases the risk of account restrictions. By keeping the final publish step manual, Oliver maintains a human-in-the-loop that satisfies both platform expectations and basic quality control.

The 60-second review also acts as a safety valve. If Larry produces something off-brand, factually wrong, or potentially offensive, Oliver catches it before it goes live. In practice, he says this happens rarely after the first few days because the analytics feedback loop steers Larry toward content patterns that actually perform.

The Results

Oliver Henry’s numbers from the first five days with Larry:

• 500,000 views across TikTok slideshows
• $588 in direct revenue from TikTok’s creator fund and related monetization
• $4,000 from MEME tokens tied to the content’s virality
• 60 seconds/day of human effort

Those numbers caught attention. Other OpenClaw users started replicating the loop.

Ernesto Lopez adapted the Larry Loop strategy for his own content operation and scaled it to $70,000 MRR. His implementation went beyond a single TikTok account, applying the same create-post-analyze-iterate cycle across multiple accounts and content verticals.

Another user reported that their Larry Loop agent hit 2 million views in 2 weeks. The specific niche and monetization details were not shared publicly, but the view count was verified through screenshots posted to the OpenClaw community.

These are real results, but they are also the best-case results that get shared publicly. Survivorship bias applies. For every Oliver or Ernesto, there are users whose loops produced mediocre content that got 200 views. The strategy works, but it is not a guarantee. For more examples of what is working and what is not, see our roundup of real OpenClaw use cases.

The Risks You Should Know About

The Larry Loop is not risk-free. Before you invest time building one, understand what can go wrong.

Platform bans. TikTok’s stance on AI-generated content and automated posting is evolving. Today’s acceptable practice could be tomorrow’s terms of service violation. If your account gets banned, your loop and its accumulated analytics knowledge become worthless on that platform. Diversifying across platforms (YouTube Shorts, Instagram Reels) reduces this risk but does not eliminate it.

Content quality floors. AI-generated slideshow content has a ceiling. It works well for informational, trend-driven, and meme-adjacent content. It works poorly for content that requires genuine personal experience, emotional depth, or complex storytelling. If your niche demands authenticity, a content agent will produce slop that damages your brand more than it helps.

Over-reliance on one channel. $70K MRR sounds incredible until TikTok changes its algorithm, its creator fund payout structure, or its policies on AI content. Building an entire revenue stream on a single platform controlled by a single company is a fragile strategy. The smartest Larry Loop operators are using TikTok views as a top-of-funnel driver and converting that attention into assets they own: email lists, communities, product sales.

API and cost considerations. Running Larry requires AI model API calls for image generation and text generation. Depending on your volume, this can cost $50 to $500/month in API usage. Make sure the math works before you scale. Our OpenClaw costs guide breaks down what to expect.

Legal gray areas. Using AI to generate content at scale and monetize it touches on evolving legal questions around disclosure, copyright, and advertising standards. Stay informed about the regulations in your jurisdiction.

Can This Work on YouTube?

Yes. The Larry Loop pattern is platform-agnostic. The cycle of create, post, read analytics, and iterate works anywhere you can access performance data programmatically.

YouTube Shorts is the most natural adaptation. The vertical format is similar to TikTok, and YouTube’s analytics API is well-documented. Some users in the OpenClaw community have already adapted their loops for Shorts with comparable results.

The key difference is that YouTube rewards longer watch time and subscriber conversions more heavily than TikTok does. If you adapt the loop for YouTube, adjust your agent’s optimization targets accordingly. Instead of optimizing purely for views, weight metrics like average view duration, subscriber clicks, and click-through rate on end screens.

For a deeper look at how OpenClaw agents can handle content workflows beyond social media, check out the Felix experiment for a different approach to autonomous agent work.

Is the Larry Loop Right for You?

The Larry Loop works best for people who meet three conditions: they are comfortable with OpenClaw’s basic setup, they have a TikTok account (or are willing to start one), and they are in a niche where informational slideshow content makes sense. If your content requires you to be on camera sharing personal stories, this is not the right approach.

If you are just getting started with OpenClaw and want to understand the broader landscape before diving into social media automation, start with how to make money with OpenClaw for a wider view of income paths, or explore real use cases to see what other builders are doing.

The loop is a tool. Oliver and Ernesto made it work because they combined it with good niche selection, consistent iteration, and a willingness to let the data guide their content strategy. The agent does the heavy lifting, but the human still sets the direction.

Have questions about setting up a content agent? Reach us at openclawdc@gmail.com.