All the best products from Anthropic's 'Builder Day' hackathon
The top teams won API credits and tens of thousands of dollars in prizes.
AI company Anthropic just hosted a hackathon at its San Francisco offices in partnership with VC firm Menlo Ventures.
More than 200 developers attended the first ever "Builder Day" to play with large language model Claude AI.
Bosses from both firms judged the hackers' products, selecting their favourite three efforts from around 100 entries.
The top teams won tens of thousands of dollars' worth of prizes, as well as API credits.
Plenty of other products impressed the judges, including a chatbot that helps users navigate the complexities of California's food stamp program.
The day began with talks including a fireside chat with Anthropic CEO Dario Amodei and a run-through of all the latest Claude updates from Head of Claude Relations Alex Albert.
Next came the hackathon itself. Developers built apps that used Claude AI — which has chatting, coding, vision and even computer use capabililities — over the course of a few hours.
The best products
Anthropic CEO Dario Amodei, head of Claude relations Alex Albert and technical staff member Soheil Koushan judged the best products at the end of the day, joined by Menlo Ventures principals Deedy Das and CC Gong.
The judges gave first place to an app that controlled a robotic arm, followed by an anti-captcha app and a product requirements document debater.
To build their robotics app, developers fed Claude the manual of an Amazon-purchased robotic arm to summarize. They used this as context to teach Claude how to use the arm, before prompting it with instructions like "rotate left until you find a human."
A camera attached gave Claude a stream of live images, allowing it to perform actions like grabbing a piece of lego held by a demonstrator nearby.
Second place went to an app that can spot when AI agents attempt to solve captcha verification puzzles.
Described as a "honeypot for AI agents" by AgentOps AI founder Alex Reibman, the uses Claude to build captchas that humans — but not AI Agents — can solve.
The third-place team asked Claude Haiku-powered AI agents to discuss the merits of a draft product requirements document and improve it.
The agents take on the roles of UX lead, data scientist, finance manager and CEO, before debating the PRD at hand.
A final agent then summarizes the debate and uses the conclusions to improve the PFD, automatically performing one of the jobs of a product manager.
If you think you can build something better, you may still have time to apply for another AI agent hackathon. AgentOps AI is facilitating another event this weekend (November 9 and 10) featuring talks on Anthropic, MultiOn, AgentOps, Neon — and more than 14 hours of hacking.
Prizes include $10,000 in AWS Credits and $1000 in Anthropic Credits for the first place team.
robot use is wild - first they took over software and it changed everything.
imagine what would happen when hardware gets involved 🤯
I really like your focus on practical applications of API design! Your tips have encouraged me to enhance my approach. Since I started using EchoAPI, I’ve seen a marked difference in my workflow.
As a team that is building an AI agent, particularly designed for meetings and task management , it's great to see the undeniable attention in this scene!