Show HN: Kafka, the first AI employee (NEW SOTA ON GAIA BY 20%)
Hacker News - AI
Jul 23, 2025 16:51
egrigokhan
1 views
hackernewsaidiscussion
Summary
Brainbase Labs has introduced Kafka, an AI "employee" capable of handling tasks via email, phone, and Slack, and performing real-world work such as coding and project management. Kafka achieves a state-of-the-art 77.2% on the GAIA Level 3 benchmark, thanks to a new "structured planning" algorithm that enables long-term, reliable task execution. This development marks a significant step toward practical, autonomous AI agents that can integrate seamlessly into human workflows.
Hi HN, I'm Gokhan, the founder of Brainbase Labs. Today we're releasing an early preview of our first generalist agent, Kafka. Kafka is the first AI employee, he comes with his own computer as well as his own email, phone and Slack so you can work with him just like you would with a regular employee. You can forward him emails, give him a call, tag him on Slack. We built Kafka as the basis for our other AI employees we will be releasing over the coming months. Kafka currently achieves 77.2% on the GAIA Level 3 benchmark, getting us closer to human performance at 87%. We've achieved this by creating a new type of planning algorithm called "structured planning" which allows Kafka to run very long term plans without getting sidetracked or hallucinating. Kafka can do some cool things, he can push code to AWS, direct its own commercial using Veo3 and do actual production tasks on Upwork/Fiverr. We're very keen to hear what HN thinks about Kafka, and how we can improve. Appreciate any feedba