https://video.ut0pia.org/videos/watch/aa34050a-3e66-4e7b-8446-0c251fa0cec1 The agent hit a login page, panicked, reported success anyway, and the upvote never happened. Tejas Kumar's diagnosis: not a prompt problem. A harness problem. The demo builds a browser agent on GPT-3.5 Turbo against Hacker News and layers in a harness without touching the prompt once. Guardrails cap iterations and compact context. A verify step reads the tool call history to catch the agent lying about what it did. A login handler watches the browser URL each loop and injects credentials programmatically when it hits the login page. By the end the cheap old model reliably logs in and upvotes the post. Speaker info: https://x.com/tejask, https://www.linkedin.com/in/tejasq/, https://github.com/TejasQ Wed, 20 May 2026 23:04:24 GMT https://validator.w3.org/feed/docs/rss2.html PeerTube - https://video.ut0pia.org https://video.ut0pia.org/lazy-static/avatars/0287a09a-aae7-4840-9843-b416426e7046.webp https://video.ut0pia.org/videos/watch/aa34050a-3e66-4e7b-8446-0c251fa0cec1 All rights reserved, unless otherwise specified in the terms specified at https://video.ut0pia.org/about and potential licenses granted by each content's rightholder.