Agent harnesses should match action and context complexity, not copy Claude Code
hugobowne · x · 2026-07-22
A long post argues that most agent systems fail because builders add too much harness complexity too early.
- The author says the common default list — memory, compaction, sub-agents, orchestration, planning, hooks, guardrails, evals — is often the wrong starting point.
- Instead, an agent should first be mapped on two axes:
- Action complexity — how many tools, decisions, dependencies, and iterations it must coordinate.
- Context complexity — how much information must stay coherent across model calls.
- Coding agents and deep-research agents tend to be high on both axes.
- Support/sales agents usually have lower context complexity; their main issues are routing, tool exposure, and human handoff.
- The post also separates action complexity from action risk: a one-turn agent can still need strict permissions if its actions have high consequences.
- The takeaway is to design around the actual failures your system can encounter, not around a one-size-fits-all Claude Code-style harness.
Related event: Production Agent Insights: Simple Architecture Beats Over-Engineering(4 posts)→
More from coding & agent
- Team-level AI agents: where should shared context and history live? — Al_Grigor · 2026-09-11
- Trust layer for money-moving AI agents: out-of-mandate actions can't get signed — Arpitbuilds · 2026-09-11
- Chaining dependent MCP tool calls: no rollback, duplicate risk — agentrsdg · 2026-09-11
- DeepMind-led paper makes design docs the source of truth, code disposable — SMART regenerates in 1.5-3h for ~$100 — Roger_M_Taylor · 2026-09-11
- Agent-built classifier labels 192k docs for $0.70 vs $13-26 with frontier LLMs — vanstriendaniel · 2026-09-11
- MathModelAgent gains traction: auto-solves math modeling and writes a submission-ready paper — jihe520 · 2026-09-11