Claude Code tiered workflow: Haiku swarms, Sonnet builds, Opus reviews at 3 checkpoints
Arindam_1729 · x · 2026-10-12
Arindam1729 shares a tiered multi-agent setup for Claude Code, arguing most people burn Opus tokens on work Haiku could handle in parallel:
- Roles: Sonnet runs the main session (writes code, executes tests); Haiku subagents search files, scan code, and fetch docs in parallel; Opus reviews architecture, catches regressions, and audits the final diff.
- Setup: CLAUDECODESUBAGENTMODEL=haiku claude --model sonnet
- 3 Opus checkpoints: before locking the implementation plan, when the same test fails twice, and before declaring the task done.
- Isolated subagents keep search results and logs out of the main context, saving large amounts of tokens.
The author also wrote a piece on running coding agents in YOLO mode with a 5-layer safety setup.
More from coding & agent
- 50B+ Tokens Later: A Practitioner's 6-Step Path Into Applied AI — brandon_galang · 2026-10-12
- Grok bot turned into a smart crypto wallet: auto-generates Solana wallet and trades — nima_owji · 2026-10-12
- One week of optimization lifts an MCP server's Smithery quality score from 73 to 98 — aniketmaurya · 2026-10-12
- DeepSeek-V4.1-Flash on 2x DGX Spark: TP2 Patches Cut First Token From 33.8s to 4.9s — rez0__ · 2026-10-12
- Running Two AI Agents on Raspberry Pis with Full Sudo, RTL-SDRs and Starlink — natesiggard · 2026-10-12
- Indie Dev Builds Kids' Book Store Almost Entirely With an AI Agent — ConstructionIcy5855 · 2026-10-12