Why 99% Accurate Browser Agents Still Fail — Derek Meegan, Browserbase
A browser agent that's 99% accurate per step succeeds only about 36% of the time over 100 steps. Here's how to beat that math. Derek Meegan, software engineer at Browserbase, explains what separates a browser-agent demo from thousands of unattended production runs. He covers how agents see the web (text, screenshots and code), why cost builds up with every step while value only comes at the end, and why you should measure success per transaction, not per run. Then he walks through a real architecture for automating a health insurance portal: deterministic tools, OCR verification, pulled-out authentication and skills that keep the agent on the critical path. In this talk: • The three ways agents interact with the browser • Why compounding step risk kills long trajectories, and how retries change the picture • Performance first, then cost and maintainability as engineering problems • A production architecture that takes everything deterministic out of the model's hands SPEAKER Derek Meegan, Software Engineer, Browserbase LinkedIn: https://www.linkedin.com/in/derekmeegan/ X: https://x.com/derekmeegan Website: https://www.derekmeegan.com/ LINKS Browserbase: https://www.browserbase.com Stagehand: https://www.browserbase.com/stagehand Stagehand (GitHub): https://github.com/browserbase/stagehand CHAPTERS 0:00 Intro 0:12 Demo vs production 0:47 Agenda 1:27 How agents interact with the web 2:32 Text, screenshots or code 3:17 Types of browser trajectories 4:32 No partial credit 5:22 The 99% → 36% problem 6:12 Profit = revenue − cost 7:37 Performance, cost, maintainability 8:12 What success looks like 8:42 Per-run vs per-transaction success 9:32 Breaking down cost 10:11 Maintenance over time 10:46 Risks: the web fights back 12:06 Architecture: automating an insurance portal 13:01 A deterministic download tool 13:31 Verifying with OCR 13:56 Pulling out authentication 14:36 Skills as standard operating procedures 15:26 The final system 16:21 Welcome to production Recorded at the AI Engineer World's Fair 2026 in San Francisco. Subscribe for more talks from the engineers building with AI. AI Engineer: https://ai.engineer YouTube: https://www.youtube.com/@aiDotEngineer X: https://x.com/aiDotEngineer LinkedIn: https://www.linkedin.com/company/aidotengineer/ #BrowserAgents #AIAgents #AIEngineer




Join the discussion
Sign in to join the discussion
Sign in