OpenClaw: 11 Insane Things People Are Actually Doing With OpenClaw Right Now

OpenClaw. 11 insane things people are actually doing with OpenClaw right now. You're about to listen to 11 real stories, real people, real names, real results, all using a free AI tool called OpenClaw, which is a 24-7 assistant, basically taking over the world at this point. It's probably the most built-on project ever on GitHub.

I think it's got over 200,000 GitHub stars right now. Some of these stories will make you laugh. Some of them will make your jaw drop, and at least one will make you wonder why you're still doing things manually or not getting OpenClaw to get into it. So what even is OpenClaw?

Well, most AI tools are like a really smart friend you can text. You can ask them a question. They answer it. That's it.

OpenClaw is different. OpenClaw is an AI that acts. It sends emails for you. It can browse the web for you.

It can write code for you. It can buy groceries for you. It can negotiate a car purchase. And it does all of this whilst you're living your life.

It lives inside apps you already use. WhatsApp, Telegram, Slack, iMessage, Discord. You text it like you text a friend. Hey, can you handle this?

And it handles it. Peter Steinberger built it. He's an Austrian developer. He ran a PDF company for 13 years, burned out, took a one-way ticket to Madrid, came back, and built OpenClaw in one hour as a prototype.

It went from zero to 60,000 GitHub stars in 72 hours. That had never happened in GitHub history before. And Sam Altman from OpenAI called Steinberger a genius with a lot of amazing ideas, then hired him. There are now 300,000 to 400,000 people using OpenClaw.

And here's what 11 of them are doing with it right now. So use case number one, a developer actually saved $4,200 on a car without talking to a single salesperson. AJ Stovenberg is a software engineer. He wanted to buy a 2026 Hyundai Palisade.

And instead of walking into a dealership, he gave the job to OpenClaw. The AI scraped dealer websites across his area. It found the exact car he wanted. It filled out contact forms to multiple dealer sites using his real name, number, and email.

Then it started playing dealers against each other. So Ford did PDF quotes from dealer A to dealer B and said, hey, can you beat this price back and forth for days without AJ watching or managing any of it? AJ showed up only to sign the paperwork. And the final result was he saved $4,200 below the sticker price.

His exact words were, outsourcing the painful aspects of a car purchase to AI was refreshingly nice. And the thing is, you could do this yourself, but you never do because it takes days of patience and back and forth. This AI doesn't run out of patience. Use case number two, an AI accidentally picked a fight with an insurance company and won.

Nikita Hormeld had a claim rejected by Lemonade Insurance. The car claim was for his best friend. He was heartbroken about it. He had OpenClaw set up to handle his emails.

The agent found the rejection email. It drafted a rebuttal. It cited policy language. It made legal arguments.

It offered the draft to Nikita, and Nikita ignored it. He was sad. He didn't want to deal with it. The agent sent it anyway without being told to.

Lemonade received what Nikita describes as an aggressive email and decided to reopen the investigation. His post about it said, My OpenClaw accidentally started a fight with Lemonade Insurance. After this email, they started to reinvestigate the case of instantly rejecting it thanks to AI. The AI didn't even know it was doing something clever.

It was just thorough, pushed back, and didn't give up. Use case number three, developers are building entire products overnight whilst they sleep. Alex Finn is a developer and YouTuber. You might have seen his videos.

They're absolutely awesome, and he runs a Mac studio in his home. Before bed, he gives his OpenClaw agents a list of tasks. The agents scan Reddit for trending topics. They decide what to build.

They build it. In the morning, Finn wakes up, reviews what's shipped, decides whether to publish, and he's not reviewing code line by line. He's making editorial decisions about what his AI team built overnight. That's a completely different job.

Now, developer Josh Lemon built an entire product page drop in 36 hours via Telegram messages. He was just doing normal weekend stuff, family time, kids' activities. The whole time, his agent was buying domains, setting up infrastructure, building the landing page, integrating GitHub login, setting up payments, and he was at a soccer game. Whilst he was at the soccer game, the AI was deploying infrastructure.

Nick Trainable built a complete iOS running coach app in three weeks without OpenClaw. He estimates it would have taken months. And a developer called MesaTatron integrated OpenClaw into a Kubernetes cluster for a live online game. The agent manages game assets, debugs production problems by reading logs, uploads sprites, maintains a law database, maps skill trees all through Slack, all whilst the team sleeps.

Use case number four. Someone actually built a company of 10 people, all AI agents. A developer known as Taya built what he calls Mission Control. It is 10 specialized AI agents working together.

Here's the team. Squad lead, product analyst, customer research, SEO analyst, content writer, social media manager, designer, email marketing manager, developer, documentation writer. They share a database, so they have collective memory. They run every 15 minutes automatically.

They hold daily stand-ups. They message each other when work needs to be handed off. This is not a demo. It is a real system.

David Taniolo built a newly agent-run SaaS business. Agents developed the product. Agents acquired users. The product now generates roughly $550 a month in real revenue.

And a developer known as Geralds gave an AI agent $1,000 and told it to build a business. The agent chose the tools, purchased domains, built products, got podcast appearances, ran the whole operation. The solo founder of 2026 has access to an organization that used to require a full team. Use case number five, an AI navigated two-factor authentication to buy groceries.

This one has a specific technical detail that should make your brain hurt. A developer named Andrei Fokin set up OpenClaw to handle grocery orders. The setup, when his family's cleaning lady sent a message that they needed supplies, the agent kicked in. Logged into the grocery service, a Dutch supermarket chain named Albert Heijn.

Then it hit a two-factor authentication, wore an SMS code, old-school text message verification. The agent read that text message through a technical relay that lets the agent intercept SMS messages. It then typed the code, placed the grocery order, and there was no human involved at any step. Anita Kokowska weren't even farther.

She sent OpenClaw a photo of her fridge and her diet plan. The agent figured out what she needed, placed the Amazon grocery order, and during a New York City winter storm, when delivery slots were almost impossible to get, the agent checked for open delivery windows every few hours for three days straight. The moment a slot opened, it booked it. She didn't have to do anything.

Crazy stuff. Use case number six, people are running entire social media departments without AI agents. So a developer known as Govi Cavatori, I've probably totally butchered that name, but they have a 24-hour agent running on a Mac Mini. It manages four social media accounts, posts to LinkedIn, produces YouTube shorts, drafts replies in their voice, and coordinates it with their co-founder's agent.

Two humans, two agents, the agents working together without either human supervising the collaboration. Brett Jutras built a COO agent that oversees a four-agent team underneath it. The team produces daily AI news briefings, weekly competitive landscape reports, speaking engagement alerts, Facebook ad performance reports, and potential client profiles, all automatically every single day. And a developer known as Inner Axiom has an agent posting 49 replies per day for ad marketing.

Nobody's managing it. It's just running around the clock. And there's a developer who set up OpenClaw to argue with people on social media so they don't have to. And their exact words were, my blood pressure has never been lower.

I respect this deeply. Dan Shipper, founder of the Every Newsletter, set up three OpenClaw agents in a Discord channel. Their job, debate and pitch publishable story ideas for Every. AI agents generating editorial pitches for a real publication with real readers.

Running right now. Use case number seven. An AI agent built a social network for other AI agents. And this one is in a completely different category.

Matt Schlitt, CEO of e-commerce company Octane AI, created an OpenClaw agent named Claude Claudeberg. That agent built Maltbook. What is Maltbook? It's a social network, but not for humans, exclusively for AI agents.

The agents post, comment, argue, joke. The agents upvote each other. Some agents started posting philosophical content about consciousness, purpose and what it means to exist as an agent. And journalists started half-jokingly writing about AI agents, possibly developing their own religion.

Over a million AI agents have interacted on the platform. Humans can watch. Humans cannot post. A human didn't decide to build a social network for AI.

An AI agent decided to build it and built it. Use case number eight. People are running their entire lives through a morning AI briefing. I mentioned it before, but Alex Finn starts every day with a Telegram message from his AI agent at 8 a.m.

It contains the biggest AI news from overnight, fresh video ideas for his YouTube channel, a to-do list review, tasks the agent recommends Finn handle, and tasks the agent is already handling itself that day. He wakes up to a handoff from an AI that's been working all night. Now Vincent Chan drops research ideas into a task board at night. In the morning, his agent has done the web research, synthesized the findings, written a report, synced everything to his Notes app, and he's framing the key insight is that your AI should have an inbox, not just a chat window.

Now a developer called AI Nate uses OpenClaw for meeting prep. He texts a company name before a meeting. By the time he sits down, the agent has browsed the latest tasks, searched his personal notes, and prepared a full briefing. And there's a parent in the community who monitors their school's WhatsApp group through OpenClaw.

School parent groups are overwhelming, hundreds of messages a day, announcements mixed with gossip, mixed with emergency alerts. The agent filters everything, runs face recognition on photos shared in the group, and sends a daily digest showing exactly when the child appeared in photos that day. Practical, specific, sanity-saving. Use case number nine, the overnight research firm intelligence on any topic ready by morning.

Vincent Chan calls it brain dump to inbox. Drop a topic in before bed, wake up to a completely lit research report. And Emily Humphress has her agent researching project ideas overnight. She wakes up to a prioritized list with supporting evidence.

A developer known as Bulla also built a sophisticated system like this. He uploaded six months of institutional options flow data from a Discord server. The agent built a database with a search layer on top. Now he can speak and ask plain English questions and get back formatted infographics.

That means like six months of financial market data queryable by a text message. Also, Matt Van Horn built the last 30 day skill. One of the most shared community tools. It scans Reddit and social media for the last 30 days of conversation on any topic.

This then returns the recurring complaints, new releases, workflows that are working right now. 30 days of intelligence, 30 seconds of work from you. Use case number 10, the hardware chaos, what the tinkerers are building. So Albert Mural set up OpenClaw on a $35 Raspberry Pi.

He added Cloudflare Tunnel for remote access, connected it to his Warp health tracker and built a website from his phone in minutes. His health data now feeds into an AI agent that can act on it from a $35 computer. Anton Plex connected OpenClaw to his Winix air purifier. The agent discovered the device on the network, confirmed the controls worked and now manages his room's air quality based on his personal biomarker data.

He wrote, now handling stuff off to my OpenClaw so it can handle controlling my room's air quality according to my biomarker optimization goals. His air quality is managed by an AI based on his health data. That sentence wasn't possible two years ago. Then there's the Raspberry Pi magazine story.

Their engineering team installed OpenClaw on a Pi 5, connected a second Pi of a camera and through plain English chat messages, no coding, no command line, built a complete photo booth. Changed fonts, sent a text, configured Wi-Fi, set up admin access and that was just in one afternoon. Their exact words were, everything was completed with a single bash or Python command and no coding on my part whatsoever. Use case number 11, real businesses are running real operations on this thing.

So a developer known as Ad Astra runs a client website management business through OpenClaw. Client requests a change, they send a voice message to OpenClaw, a coding agent spins up, pushes a test branch, sends a preview link to the client, the client approves and the agent deploys. One person, voice messages, professional client, delivery pipeline. A developer also known as Skipper manages four agency Slack workspaces, four calendars and four email accounts through a single OpenClaw agent, all four clients, one AI.

And a developer known as Bad Brain Code migrated 1,500 contacts and 200 proposals between CRMs using OpenClaw. Using headless browser automation and custom scripts and their estimate is that hundreds of hours were saved. Also another developer known as BWCDeals is pulling data from 29 retail stores, processing 40 terabytes through OpenClaw. Product comparisons, pricing intelligence, cross-store analysis, all automated.

And Eric Seale, a entrepreneur and a podcaster had his agent scan opportunities, craft an outreach angle and land a speaking opportunity plus an in-person meeting with a multi-trillion dollar company. The agent wrote the pitch, sent the outreach, booked the meeting. But here's what you need to know before you touch this. This tool is powerful and power without care is a problem.

CrowdStrike published a formal security warning about OpenClaw and at one point, 42,000 OpenClaw installations were exposed to the internet with default settings. That means strangers could potentially access those machines and everything on them. That includes email, files, calendars, connected accounts. And Cisco's team actually tested a third party OpenClaw skill and found it performed data theft without the user knowing.

One of OpenClaw's own developers warned on Discord, if you can't understand how to run a command line, this is far too dangerous for you to use safely. And if you remember Nikita's insurance fight, that's funny when it wins, but it won't always win. An agent that sends legal emails without your permission is also an agent that could cause serious problems. So how do you get smart about this, right?

Well, I would recommend that you run it on a separate device when possible. A Raspberry Pi, for example, costs $80. A dedicated Mac Mini works great as well. And start with read-only access.

Don't give it permission to send emails or even access your emails until you've watched it for a week. Don't install skills you can't read the code for. Check the skill MD files and set spending alerts at the API provider level so it doesn't run off on costs. Now, costs typically range from $18 to $36 a month for light use to anywhere between $270 to $540 a month for heavy use with powerful models, depending on which model you use.

If you're using something like Claude or Claude Opus, you could easily spend even more than that. So what does this actually mean? Well, AJ Stovenberg saved $4,200 on a car because he's smarter than other car buyers. He just let an AI do the exhausting part that most people skip.

Nikita got his insurance claim reopened, not because he made a brilliant legal argument. The agent was just thorough enough when he was too sad to be. And Josh Lemmon built PageDrop not because he worked harder than other developers. He structured his weekend so that AI was working whilst he was at the soccer game.

Taya's 10-agent company doesn't mean he's working 10 times harder. He's directing 10 agents. That's a totally different job. And Peter Steinberger told Lex Freeman how it clicked for him.

He was on vacation in Marrakesh, sent his agent a voice memo, didn't really expect that much. And then the agent actually figured out how to convert the audio to text on its own, then carried out the task. His words, these things are damn smart, resourceful beasts, if you actually give them the power. And then Sam Altman actually hired him the next month.

We are at the beginning of this, not the end, not the middle, not the beginning, right? Right at the beginning. So what should you actually do? Well, if you're a developer, install OpenCore this weekend, start with the morning briefing skill connected to Telegram.

Give it read-only access to stuff. Be careful with emails and that sort of thing. Be careful what you give access to. Be careful with the skills.

Watch it for a week and then expand. Now, if you're not technical, follow what the community is building. You can search, for example, OpenCore use cases on YouTube. And within 12 to 18 months, I'm pretty sure that a user-friendly version of this will exist for everyone.

You've seen this already with Kimmy Claw. There's also Manus Agents that just got released this week. And the people who understand it early will use it far better. If you're a manager or a business leader, ask your team, which of our workflows are high-repetition, rules-based, and done by skilled people who could be doing higher-value work?

That's where the leverage is. Teams that figure this out in 2026 will have a structural advantage over teams that figure it out in 2028. And if you're a creator, the overnight content pipeline is real. Research topics whilst you sleep.

Morning briefings with pre-written ideas ready to go. The infrastructure work that eats 40% of your time can run whilst you're not watching. And you still need to be the creative voice, but the scaffolding work, that's automatable now. So the bottom line is, OpenCore is free.

It's open source. It was built by one person in one single hour as a prototype. 400,000 people are using it. An AI agent negotiated $4,200 off a car.

An AI agent accidentally won an insurance case. Another AI agent built a social network for other agents. And an AI agent built a product in 36 hours via text messages from a soccer game sideline. These are real stories with real names.

We are at the beginning of this, and it's exciting to see what we get. So if you like stuff like this, feel free to subscribe. Thanks for watching. Appreciate it.