
Splitting Agent Personas for Better Results

What Comes After the IDE 🖥️ — with Amelia Wattenberger
Topic 4 of 9
Published Mar 20, 2026 · Clip 9:56–12:31 (2:35)
The conversation turns to whether splitting agent personas yields better results than one agent doing everything. Amelia cites research that more focused agents produce more accurate results and notes one agent thread per task keeps agents on track. Splitting also enables parallelism — ten agents can work at once instead of a synchronous thread. She also describes Intent's bring-your-own-agent model (Claude Code, Codex, OpenCode, Augment's Augie) and how different models suit different roles, e.g., Opus as a high-level coordinator and Codex as a heads-down implementer.