Don't Be Fooled by Average Turns: Coding and Customer Service Agents Need Different Harnesses
hugobowne · x · 2026-07-28
The author argues that designing AI agents based on "average conversation turns" is misleading, much like the statistically true but useless fact that the average human has one testicle.
Different agents require fundamentally different harnesses:
- Coding/Deep-research agents: May run for dozens or hundreds of turns. They need to manage growing context, preserve evidence, execute and verify long action sequences, and sometimes divide work among sub-agents.
- Customer-service agents: Should often resolve requests in 1 to 5 turns. While they may not need compaction, elaborate memory, or sub-agents, they still require reliable tools, state management, guardrails, traces, and human handoffs.
The author is hosting a course to teach building AI agents from first principles, starting with a working customer-service agent and adding tools, state, context, guardrails, and MCP connections.
More from coding & agent
- Controlled study finds multi-agent setups give 0.0% average gain over a single agent — bravo_abad · 2026-07-28
- Persistent state for coding agents should include approvals, tool traces, and validation results — DesktopLabHQ · 2026-07-28
- Codex Computer Use is showing obvious PMF in compressed computer workdays — hudzah · 2026-07-28
- Splitting planner and executor roles cut my agent context bloat and token bills — truecakesnake · 2026-07-28
- GitHub repo shows how Streamlit theme config helps agent-assisted UI edits — andfanilo · 2026-07-28
- Bilinc launches a hosted MCP memory server to help agents remember across sessions — atakanelik34 · 2026-07-28