AI Engineering Workflows, Benchmarks, and the Future of Coding Agents | swyx

AI engineering workflows are evolving fast. swyx (AI.Engineer) breaks down agentic code benchmarks, why 50% of SweBench code is unmergeable, how agents are breaking GitHub, and why all the AI labs are IPO-ing at the same time. Chapters below. Subscribe for more AI:AM conversations. 0:00 AI insiders are selling. 0:27 $300K vs $5M for same job. 0:49 OpenAI is the Apple of AI. 1:37 Intro: Who is swyx 4:04 AI Engineer World's Fair themes 7:24 Continual learning: Weights vs systems 10:51 Enterprise AI: Cheap, perfect, private 14:45 Startups vs enterprises: Capability vs cost 17:38 FrontierCode: A new AI coding benchmark 23:15 Preventing benchmark saturation 25:43 Slop code, human taste, and Move 37 30:13 Claude Opus vs Fable: Cost vs capability 32:05 The advisor model and model routing 36:28 Convergence and market segments in AI 44:15 Rebuilding cloud infrastructure for agents 51:46 Vibe coding internal SaaS replacements 57:22 Whoever owns the system of record wins 59:55 The AI IPO bubble and insider selling 1:04:49 Solving Star Trek problems after the IPO 1:14:07 Career advice for CS grads in the AI era 1:19:50 AI Engineer World's Fair 2026 Full episode:    • swyx: How AI Agents Are Rewriting Software   Guest links: swyx: https://x.com/swyx |   / shawnswyxwang   AI:AM links: 𝕏 AI:AM: https://x.com/ai_in_the_am 𝕏 Prakash: https://x.com/8teapi 𝕏 Nathan: https://x.com/labenz Website: https://ai-in-the-am.com #AIEngineering #CodingAgents #AIBenchmarks