Anthropic, Browser Use and KERNEL

The State of Computer Use Agents | Anthropic, Browser Use & KERNEL

Bot traffic on the internet just passed human traffic, two years ahead of forecast. Most of it is agents clicking through websites built for people.

So I got three of the people building those agents in a room: Lucas Gonzalez Pagliere, who works on computer use at Anthropic, the team that shipped the first computer use model back in 2024. Reagan Hsu, founding engineer at Browser Use, whose open source library is sitting at around 100K GitHub stars. And Eric Feng, founding customer engineer at KERNEL, which runs the browser infrastructure underneath a lot of this - he was also first GTM at Sentry.

We covered where the models genuinely are today vs where the benchmarks say they are, what it costs to run an agent long enough to finish real work, and the arms race between agents and the anti-bot systems trying to keep them out. Then the harder question underneath all of it: whether the web reorganizes itself around agents, or hardens against them. They disagreed on plenty of it.

If you want to know what agents can actually pull off on a real website today - and what still stops them cold - this one's worth your time!

Episode01:06:40
Watch on YouTube
The State of Computer Use Agents | Anthropic, Browser Use & KERNEL
Episode still: Anthropic, Browser Use and KERNEL

Chapters

  1. 00:00Welcome and Guests
  2. 00:49Companies and Stacks
  3. 01:14Everyday Agent Use Cases
  4. 03:38Defining Computer Use Agents
  5. 05:15How Computer Use Works
  6. 08:41Screenshots vs DOM Hybrid
  7. 11:03Benchmarks and OSWorld
  8. 13:51OSWorld 2 Difficulty Jump
  9. 15:43Training Models and Cost
  10. 18:57Speed Infrastructure and Stealth
  11. 21:55Anti Bot and KYC Future
  12. 27:24Reverse Engineering vs UI Automation
  13. 29:43Computer Use vs Browser Use
  14. 31:05Scaling Laws and Harnesses
  15. 32:49Playwright Selenium Still Matter
  16. 33:17Playwright Still Dominates
  17. 33:27Why Run 1000 Agents
  18. 34:39Long Running Agent Challenges
  19. 35:45Memory and Compaction
  20. 38:10State Changes Mid Task
  21. 39:36OS and Browser Fingerprints
  22. 40:42DOM Efficiency and WebMCP
  23. 41:36Recsys and Agent Personas
  24. 43:57Agent Friendly Websites
  25. 45:41Human Speed vs Agent Power
  26. 48:32Context Window Tradeoffs
  27. 50:34Harnesses for Temporal State
  28. 51:59Speed Optimizations and Tabs
  29. 54:23Human Collaboration Limits
  30. 58:09Raw Capability vs Better APIs
  31. 01:00:47Training Methods and Bottlenecks
  32. 01:01:41End State Interfaces
  33. 01:05:09Next 12 Months Predictions
  34. 01:06:17Closing Thanks

Related topics

More episodes