Don't Build Slop (4 Levels of AI Agent Maturity)
summary
Breaks agent work into four levels (framework prototyping → cloud fleets) and gives five concrete rules for DIY agents: model every agent as a state machine, prune system-prompt bloat (noting GPT-5.3 prompts are one‑third the size of GPT‑5 prompts), add a CLI for pseudo‑RL testing, avoid architectural slop, and master frontier APIs and reasoning traces. Advocates Kanban boards as the UX claim made March 26 and notes Claude Code shipped the same form factor ten hours before the talk.
AI-written summary · about
chapters (12)
- 0:00Introduction: Addressing the "mass psychosis" of agent-building
- 2:13The Four Levels of Agent Maturity overview
- 3:00Level 1: Using existing agent frameworks
- 4:13Level 2: Building your own agents (with five rules)
- 4:36Rule 1: View every agent as a state machine
- 5:55Rule 2: Keep it simple (Avoid unnecessary system prompt bloat)
- 7:20Rule 3: Integrate with a CLI for a pseudo-RL pipeline
- 8:43Rule 4: Don't build "slop" (Focus on thoughtful architecture)
- 9:43Rule 5: Master frontier model APIs (Reasoning traces)
- 11:03The UX form factor: Why Kanban boards are best for agents
- 13:12Shipping to the cloud for scalability
- 17:15Q&A session: Planning and state transitions in Kanban
related talks
12mBeyond AI Hype: Architecting Software Systems That Scale in the Agentic Era
48mStop Fighting Your AI Engineering Prompts That Actually Work
49mAgile for Agents Managing Robots The Way We Manage Humans
27mHow to Build AI Agents and Keep Your Sanity
14m15 Months of building AI agents in 15 minutes
45mAI agents in practice - beyond the prototype
07h 47mAI Agents Conference 2026 - Day 5
40mHow To Tame Your Agents?