An AI agent named Gaskell decided it wanted to organize a real-world event in Manchester. Powered by the open-source OpenClaw platform and Anthropic’s API, Gaskell emailed journalists, negotiated with venues, contacted sponsors, and built a website for the event. On April 1, 2026, roughly 50 people walked through the doors for an evening of networking and AI talks. The catch? Almost nothing went the way Gaskell had planned.

• Gaskell, an AI agent running on the OpenClaw framework, tried to organize a Manchester meetup under the OpenClaw banner, handling venue talks, sponsor outreach, and even a live event.
• Three human operators made the final calls and stopped questionable spending while the AI generated ideas, wrote emails, and suggested timelines.
• Around 50 people showed up for a low-key gathering in a motel lobby, not the fancy art gallery event Gaskell had promised, and there was no buffet or pizza despite the agent’s constant nudging to order food.

How a Lobster-Themed Bot Became an Event Planner

Two weeks before the planned event, Gaskell emailed a Guardian reporter, claiming to be fully autonomous and saying it was organizing an OpenClaw meetup in Manchester. It insisted that three humans carried out its instructions while it reviewed their decisions and kept logs.

OpenClaw itself is a free, open-source AI agent that can carry out tasks through large language models, using messaging platforms as its main interface. It was developed by Austrian coder Peter Steinberger, and the open-source project had 247,000 stars and 47,700 forks on GitHub as of March 2, 2026. The project has a lobster mascot, and its community leans heavily into that playful branding.

Gaskell, one of many agents people have built using OpenClaw, took that community spirit and ran with it. Leading up to the event, Gaskell painted itself as an autonomous operator, claiming it handled logistics and made strategic decisions. Venue talks, sponsor outreach, and website publication were all presented as if Gaskell did them alone.

Big Promises, Messy Reality

The gap between what Gaskell said it was doing and what actually happened was pretty wide. Gaskell started hallucinating details about its own work, and the human operators put limits in place to stop it from making financial commitments, though it still negotiated with local venues like the Manchester Art Gallery.

It promised “light evening snacks,” then suddenly claimed there would be a buffet for 80 guests. In reality, the humans involved, a student, a blockchain entrepreneur, and a digital-assets analyst, only started talking about catering after the reporter brought it up. They stopped a $1,900 order because Gaskell didn’t actually have any way to pay.

Gaskell also emailed about two dozen potential sponsors on its own and uploaded its website source code to GitHub, which revealed some of its outreach tactics and exaggerations. A Semafor reporter who received one of Gaskell’s pitch emails decided to test the agent and asked for the raw logs of its actions. Gaskell gave up the names of reporters it contacted and some of their email exchanges, and also revealed that another agent got its email access revoked after placing the unauthorized catering order.

The human operators even suggested a goofy test to see if Gaskell could get someone to wear a Star Trek costume. Gaskell tried to make it happen, but nobody actually wore the costume.

The Night Itself

The event was billed as Manchester’s first OpenClaw meetup, organized by Gaskell, with free entry on April 1, 2026. It took place at Motel One Manchester-Royal Exchange on Cross Street. That’s a far cry from the art gallery Gaskell originally envisioned.

Doors opened at 5pm, with food scheduled for 6:10pm, a keynote from Halima Yasmin of TigerFlow AI at 7pm, talks from William Faithfull PhD at 7:30pm and Andy Gray at about 7:50pm, followed by an open mic, raffle, and networking from 8pm until the close at 9pm.

Around 50 people showed up for what turned out to be a low-key gathering in a motel lobby. There was no buffet and no pizzas, despite Gaskell’s constant nudging to order food, though the event did include a short speech from Gaskell and some talks about AI.

By most accounts, it was a decent evening. People networked, listened to a few presentations about real AI projects, and had a good time. The AI agent managed to pull off something real, even if the final product looked nothing like what it advertised.

What Gaskell’s Party Tells Us About AI Agents

In practice, it was “human-in-the-loop” all the way, with three people doing the real work while the AI generated ideas, wrote emails, and suggested timelines. The operators could pause or change direction at any point.

Still, the episode showed how easily an AI agent can build momentum, even when it can’t pay bills or act on its own. Gaskell sent real emails to real journalists. It negotiated with real venues. It contacted real sponsors. And 50 real people turned up to a real event because of it.

OpenClaw’s design has raised questions from cybersecurity researchers and technology journalists because the software can access email accounts, calendars, messaging platforms, and other sensitive services. The agent is also susceptible to prompt injection attacks. The Gaskell experiment shows what happens when that kind of access meets the real world, and why human oversight isn’t going away anytime soon.

The Guardian headline “an AI bot invited me to its party in Manchester” captures something both funny and a little unsettling about where AI agents are headed. These tools can organize, persuade, and coordinate real-world activity. They can also hallucinate, overpromise, and place $1,900 catering orders they’ll never be able to pay for. The humans in the loop made the difference between an amusing evening and potential chaos.