Claude Is Starting to Improve Itself (Not How You Think)
Fiddler The Ideal Partner for Agents - https://www.telerik.com/fiddler/fiddler-everywhere/mcp?utm_source=robshocks Software is starting to improve itself, just not in the sci-fi way. In this video I look at the loop that's doing the improving: every correction gets checked and written into the system, so your coding agents don't need it next time. I lay out a six-rung ladder for where self-improvement fits, then go through how Addy Osmani (AGENTS.md and reflections), Lauren Tan (pstack's /correct), Matt Pocock (/retro) and Zach Lloyd (skills that improve skills) each close the loop. Then Patrick Collison's liquid software, which puts the loop inside the app itself, and where I draw the line between all this and the recursive self-improvement headlines, including METR's new assessment of AI speeding up AI research at Anthropic. Then where to start: guardrails and verification, skills in your repo, evals, and a retro step. This video is sponsored by Fiddler. Connect Fiddler Everywhere's MCP server to Claude Code, Cursor or Codex so your agent can see the real requests and responses. Free 10-day trial: [TRACKED FIDDLER LINK] COURSE & COMMUNITY https://switchdim.s.gy/yt-self-improving-software SOCIALS X: https://x.com/robshocks LinkedIn: https://www.linkedin.com/in/robertoshaughnessy/ Instagram: https://www.instagram.com/therobshocks/ RELATED My pstack video: https://www.youtube.com/watch?v=lUhXa8GiXns SOURCES Addy Osmani, Self-Improving Coding Agents: https://addyosmani.com/blog/self-improving-agents/ Addy Osmani, Loop Engineering: https://addyosmani.com/blog/loop-engineering/ Lauren Tan, pstack: https://github.com/cursor/plugins/tree/main/pstack Matt Pocock, retro skill: https://github.com/mattpocock/skills/blob/main/skills/engineering/retro/SKILL.md Zach Lloyd, Don't vibe code your software factory: https://www.warp.dev/blog/build-production-software-factory Patrick Collison, Liquid software: https://patrickcollison.com/dispatches/liquid-software METR, AI R&D acceleration at Anthropic: https://metr.org/blog/2026-11-05-anthropic-ai-rd-acceleration-assessment/ WORK & COLLABS info@switchdimension.com CHAPTERS 0:00 Intro 1:14 The Loop, Not the Code 2:47 Sponsor: Fiddler 3:54 The Ladder 5:04 Write It Down 6:02 Make It Impossible 8:11 The Outer Loop 9:16 Liquid Software 10:33 Where I Draw the Line 11:34 What to Build First 11:57 Outro #ClaudeCode #AIAgents #SoftwareFactory #SelfImprovingSoftware #RecursiveSelfImprovement #LiquidSoftware #AgentSkills #AgentMemory #LoopEngineering #METR #CodingAgents #AIDevelopment #CursorAI #AgentDevelopment
