Don't Build Slop (4 Levels of AI Agent Maturity) - Ara Khan, Cline

The prompt for GPT-5.3 is one-third the size of the one written for GPT-5. Frontier models are so capable that longer system prompts cause sensory overload and degrade performance. The rule Ara Khan keeps returning to: every single thing you add to an agent risks making it worse. The talk breaks agent-building into four levels, from framework prototyping to cloud-native fleets, with five concrete rules for writing your own agent code in between. The form factor argument lands on Kanban boards as the right UX for managing parallel inference-bound agents. He made that prediction publicly on March 26. Claude Code shipped the same thing ten hours before this talk. Speaker info: - https://x.com/arafatkatze - https://www.linkedin.com/in/arafatkatze/ - https://github.com/arafatkatze