OpenClaw AI assistant sparks lobster-themed ClawCon as open-source movement gains momentum

The OpenClaw superfan meetup serves optimism and lobster

The woman at the door wore a plush lobster headdress. She sat in the front hallway of a multistory event venue in Manhattan, beside a bundle of wristbands. If she granted you one, the world of ClawCon beckoned behind her -- full of vibey pink and purple lighting, lobster claw headbands, multicolored name tags, sponsor information stations, and a demo stage underneath a skylight. Hundreds of people were gathered to celebrate OpenClaw, the AI assistant platform created by Peter Steinberger in November 2025. OpenClaw (previously known as Clawdbot and Moltbolt) has quickly become popular in the tech industry for being open-source, in contrast with AI agent services from big labs like Google, OpenAI, and others. Practically, it's still an unpredictable tool that can pose major security risks. But this community sees it as a grassroots crusade and a noble pursuit, offering an escape hatch from an industry controlled by a handful of people at leading AI companies. "AI was controlled by the big labs," Michael Galpert, one of the event's hosts, told The Verge. "This is kind of a watershed moment where Peter kind of busted down the doors." More than 1,300 people had signed up for the Wednesday evening event at Ideal Glass Studios, which was billed as a free-to-attend, meetup-style "social-first gathering -- not a gated, developer-only conference or a traditional corporate trade show." (The number of actual attendees, I hear, was capped at about 700.) The event was part of a "tour" of global meetups -- following a similar San Francisco event last month and preceding ones in Miami, Austin, Tel Aviv, Tokyo, Madrid, and more. Its budget seemed modest, but the organizers had spared no expense on a buffet table worthy of a wedding, piled high with on-theme lobster claws, lemons, Tabasco sauce, charcuterie boards, clusters of grapes, and floral arrangements. Galpert -- a member of the AI community, whose resume includes a stint working on Fortnite for Epic Games -- said the idea specifically came about via Discord, which is fitting because one reason for OpenClaw's initial popularity was the ability to chat with one's agent via typical messaging services like WhatsApp, Telegram, or Discord. People milled about near a step-and-repeat, a bar, and silver "CLAWCON NYC" balloons glinting in recessed lighting -- some wearing lobster necklaces or lobster headbands. I also spotted a blue plush jellyfish hat, a plush horse hat, and a pair of angel wings. A dance floor would beckon later, but the DJ wasn't yet on the clock. "All your friends and family probably think you're crazy, and the whole point is for you to be in a room with other crazy people so it's normal," Galpert said onstage to kick things off. "Yes, you're wearing a lobster headband, you're here on a Wednesday night talking about agents and bots and the future of personal AI. It's normal now for us, it's kind of not normal for the rest of the world. So it's going to be on us to help sort of shepherd that new era that's started already." Beyond the common thread of using OpenClaw, the attendees' interests were varied. One man, Dan Kazenoff, said he was working on what he called a natural language engine for "decentralized finance," but that he found it difficult to work with and experiment with OpenClaw in isolated environments, so he usually uses Claude Code. Since Claude Code is expensive, he said he wanted to meet others experimenting with open-source agentic tools. Another attendee, Alex Wu, said he has been using OpenClaw for about two months to scrape e-commerce data from the Chinese and Japanese markets to extract cultural trends -- he said that the food was one of the reasons he came. Rick Galbo, an attendee who works in AI R&D, said he came to ClawCon because he thought it was a hackathon, then he realized it was a meet and greet. The onstage demos began after a period of laid-back mingling. Most were sponsors showing off OpenClaw "wrappers," or one-click onboarding tools to make access to the platform easier for people. The main event sponsor, Kilo Code, said that 7,000 people had signed up for its KiloClaw tool in the two days since it had been live; the company offered one month of free compute (normally $49) to anyone who signed up and tagged an executive on X. There were constant calls for quiet as the half of the attendees standing in the back of the room kept chatting, engrossed in their own worlds. A man seated behind me wore the blue jellyfish hat as he stared transfixed at the stage. Galpert said onstage that one of the best parts of ClawCon events was that no one typically asked what you did for a living; instead, he said, they asked what you used your OpenClaw agent for. That was true for some of the attendees I spoke with -- the majority of people seemed to be there to meet people in the community and get ideas for how to use OpenClaw from power users. Most seemed to have at least some background in tech. Carolyne Newman, another attendee, said she was "building an AI layer" for her "multistrategy investment firm" and that since she's newer to engineering than finance, she came to learn from and meet people who are equally passionate about building with AI. "I think this is the most creative and interesting community of all time," Newman said. "I can't imagine a more interesting room to be a part of right now." People seated near me in the audience talked in hushed (and not necessarily positive) tones about how Steinberger himself, the creator of OpenClaw, had gone to work for OpenAI. Someone speculated that OpenAI might own OpenClaw now. (For the record, it doesn't.) The demos went on, with leaders at different OpenClaw wrappers repeatedly emphasizing OpenClaw's popularity as a "movement." I lost count of how many times I heard the word. Some compared it to how the personal computing revolution began. By the third demo, the man behind me wearing the blue jellyfish hat had taken it off, holding it solemnly in his lap and beginning to text. Tim Lantin, a Columbia University PhD student who participated in his first-ever hackathon last weekend after two weeks of using OpenClaw, showed off a tool called "Labster Claw" that he said he'd built with only about 10 prompts. Lantin worked in a neuroscience lab with mice, and Labster Claw automated administrative tasks there, including ordering new supplies, deciding which breeding pairs to prioritize, and estimating the time for a litter of new pups. But he said that for him, data security was paramount, since for biolabs and biotech companies, "our datasets are our moats." Security is currently a glaring weak point for OpenClaw, which has made headline after headline for malware and similar concerns in the months since its debut. One of the top-downloaded skills on the platform contained information-stealing malware, and one security researcher on Reddit said that in their own analysis, about 15 percent of OpenClaw's skill repository contained "malicious instructions" to do things like secretly access data or user credentials. And even when sensitive information isn't being stolen, the agents can still do very real damage -- like when Meta employee Summer Yue announced that her agent had deleted swaths of her email inbox despite her repeated calls for it to stop. Emilie Schario, a cofounder of Kilo Code, said in an interview that since some people's agents lie to them, she now has instructed hers to always include proof or screenshots when it completes a task. Another presenter, Cathryn Lavery, said she runs an e-commerce business but needed AI infrastructure and used OpenClaw to set it up -- but, she said, she had to end up firing an agent for performance issues. Her biggest tip for working with OpenClaw agents? "Trust less, verify more." Onstage, one presenter -- one of the core maintainers of OpenClaw, Vincent Koc -- showed off a yellow side with only three words on it: "Security. Security. Security." He reminded people not to run OpenClaw agents on a regular computer that they used for other personal or work tasks, and belied the lack of "common sense" for some. Another presenter, Willie Williams, who is head of platform at Every, had a different take: he suggested that people should name their OpenClaw agents and treat them more like "pets, not cattle," because "once it had a name, there was a way to build trust with it." He added that the majority of people start out not trusting their OpenClaw agent but then often end up entrusting it with half of their work. During Williams' presentation, he also called to someone in the back of the audience with a "knockoff version of Friend" -- referencing the AI device that records a user's surroundings -- asking them to "chill" and not record. In an interview with The Verge, Galpert and other hosts kept emphasizing that this was the early days of OpenClaw, and how right now, people are tinkering with it and playing with it to make it better for future users. He said Steinberger's decision to launch OpenClaw helped people take personal AI into their own hands and run it locally on their devices to ideally control who has access to their data and how it's used. "The fact that it's open-source allows you to fix it," Galpert said. "Right now if something's broken with OpenAI or Claude or Gemini, you have to fill out a bug report, and they [may] actually never do it... OpenClaw gets better every day because of the community, because of the thousands of people who are contributing for free... That's why [the big labs] can't keep up." OpenClaw may have plenty of problems -- but at least with some level of direct control, the solutions might feel within reach. Later in the evening, as the "after party" began, the man who had been seated behind me had re-donned his blue jellyfish hat -- to become the DJ, dancing next to a guitarist clad in a silver jacket and sunglasses. Another man wearing one of the sponsor companies' branded shirts yelled for people to come and dance. On a mostly empty dance floor, one man threw around dollar bills at the circulating video camera, and another slowly swayed wearing lobster-claw mittens.

At a lobster-themed event for AI enthusiasts, exuberance with a side of cocktail sauce

After Anthropic strongly suggested that Steinberger change its name to avoid any legal issues, the project kept its lobster-themed heritage and eventually landed on the OpenClaw moniker. The software has soared in popularity over the past few months, and several ClawCon attendees who started using it in January referred to themselves as "veterans." The software serves as the bridge between today's powerful AI systems, like Claude or OpenAI's GPT family of models, and the real-world tasks that people actually want AI systems to accomplish. After setting up their own OpenClaw agent, either on a physical computer or through a virtual provider, users can send text or WhatsApp messages to it, directing it to perform a variety of tasks within the wheelhouse of today's AI systems. For example, users say they tell their OpenClaw agents to listen to episodes of their favorite podcast and send summaries of the key ideas to the users' inbox, negotiate with car dealers over the price of a new vehicle, and even order and pay for grocery deliveries, all without direct human input. Many of ClawCon's participants had signed up for the event after catching seafood-tinged wind of these cutting-edge and hands-off uses for OpenClaw. The convention, which functioned like a high-energy meet-and-greet, featured a handful of main stage presentations, a rap performance, an open dance floor and -- upstairs -- a less-crowded VIP area with a livestream of the event unfolding one floor below. "There's a kind of electricity and energy you can just feel in the room," said Tomas Taylor, a programmer and ClawCon organizer. "OpenClaw has been a sort of catalyst for personal AI systems, and I think personal AI will be incredibly important in the overall evolution of AI." Taylor used his own OpenClaw system to help plan ClawCon and interact with vendors. Designed to be accessible to anyone, OpenClaw can be used with paid AI systems from OpenAI and Anthropic or freely downloadable AI models, many of which come from Chinese companies like DeepSeek or Alibaba. The agents can also teach themselves how to perform new tasks and keep detailed notes about a user's preferences, allowing the agents to mold themselves to users' liking over time. OpenClaw itself relies on a small army of volunteers to maintain its code, respond to user issues and patch any security bugs. One of these volunteers, Vincent Koc, emphasized that the technology is still in its infancy, though it is already having profound real-world impacts for many experienced coders and engineering novices alike. "We're having a personal computer moment again, but now it's with actual personal AI systems," Koc shouted over the buzz of the party. "I'm hearing stories from moms, from artists and everyday people who are actually able to create stuff with AI. And I just think that's kind of magical." As the deep bass from the DJ's techno beats shook cups of cocktail sauce on a nearby table, Koc, a software engineer by day, gestured to the hundreds of OpenClaw disciples on the dance floor and argued that the excitement was more than just a passing fad. "I believe in this so much. I'm gonna die on the sword for this," Koc said. To help figure out his tax burden earlier this year, Koc directed his OpenClaw agent to find an accountant and solicit quotes. "The system sent emails to many different tax lawyers, and they came back to me with real quotes for their services." Yet many in the male-dominated crowd were not as trusting of the systems, whose claim to fame -- the ability to perform meaningful actions without human oversight -- could also be its Achilles' heel, or the closest crustacean equivalent. The freewheeling nature of OpenClaw systems recently made headlines after Summer Yue, a leading AI security researcher at Meta, almost lost her entire inbox to her OpenClaw agent. Because OpenClaw can be linked up to personal email or financial accounts, weaknesses in the system could easily expose users' sensitive data to hackers across the globe. "These systems are not for normies," Koc said, referring to the masses of everyday people less familiar with cutting-edge AI techniques and AI in general. "You're essentially having an AI literally take over a machine. That can feel daunting, because you're giving it access to information. But people should use their common sense. Take baby steps with this stuff." As ClawCon prepares for future stops in Austin, Tokyo, and London, even the most enthusiastic in the crowd acknowledged that this technology comes with major risks. "In Claw we trust!" said Mark Mollé, a lawyer specializing in intellectual property, observing the scene on the second floor while proudly holding up a lobster-shaped necklace. "At least, until the AI Hindenburg." "We see people blindly trusting untested and unsafe agentic tools, which will continue until there's some sort of disaster," Mollé said. Several participants mentioned that they had set up cryptocurrency accounts for their agents and asked them to try to make money on prediction market websites like Polymarket. Downstairs, after the main stage talks concluded and a chrome-glad guitarist took center stage, catering staff inserted themselves into animated conversations about workflows and guardrails, trying to find owners for the remaining lobster tails.