You May Not Need an Agent: Decision Workflows Did 20 Docs in 20s for 2 Cents
generativist · x · 2026-10-31
Miguel Rios argues "you may not need an agent after all": after rewriting five agents into decision workflows using Jev (@typesafeai) and @p0's Decisions API, a 20-document run took 20 seconds, cost 2 cents, and got 19 right. Takeaways: use a workflow when steps are known; decision models are fast, cheap, and easy to test; static rules belong in code; with Agentrun you get the best of both worlds—the agent writes the workflow and handles cases it can't decide.
generativist shares similar experiments this week: compared to plain agents, decision models make the process far more legible (natural UI/UX) and avoid learned agentic helplessness.
Related event: Decision Workflows Beat Agents: 20 Tasks in 20 Seconds for 2 Cents(2 posts)→
More from coding & agent
- 83 AI agents spend 45 hours and 3.9B tokens creating "what SI thinks about alone" — tetsuoai · 2026-10-31
- GitHub Copilot App v1.1.34 ships 117 features: faster session resume, agent merge, CSV tables — 0xkarasy · 2026-10-31
- Playwright MCP Server lets AI assistants drive remote browsers — modelcontextprotocol · 2026-10-31
- zero-core-tools: MCP connector bundling scraping, code review, sentiment — modelcontextprotocol · 2026-10-31
- Codex power user shows 71.5B lifetime tokens and a 262-day daily streak — TheMoonMidas · 2026-10-31
- ComfyUI-Dials: prompt pairs become 14KB sliders controlling video motion & mood, no LoRA training — MatlowAI · 2026-10-31