Fewer Tools Make Agents Stronger
Street_Inevitable_77 · reddit · 2026-07-13
While building agents, the author found that more tools do not necessarily make them stronger, but instead slow them down, increase cost, and cause more selection errors.
Key Findings
- Fewer, broader tools often work better than many fine-grained ones
- Models are good at deciding 'what' to do, not 'how'
- Merging 10 fine-grained tools into 2-3 high-level intent tools improved performance
- Return structured data, not long text
- Free-text outputs cause models to treat them as new prompt sources
- Separate factual tools from what-if/simulation tools
- Mixing them leads models to treat simulated values as real
- Tool and parameter descriptions are not documentation, but context budget
- Compressing descriptions to one line improves reliability and token cost
The author's experience: agents are good at deciding 'what' to do, tools should hide 'how', and tool surfaces should be minimized to fewer decision points.
More from coding & agent
- Agent verified 'success' after double-voting: a lesson in harness vs model bugs — TejasKumar_ · 2026-10-26
- diginomica: agentic automation hype is overdone, and 'The Hackening' of enterprise security has begun — jonerp · 2026-10-26
- 'Vibe verification': ~60% of AI-written formal specs prove the wrong property — basedjensen · 2026-10-26
- Devin launches Security suite powered by Claude Mythos with $100M+ credits — DevinAI · 2026-10-26
- HackBench injects reward hacks into agent trajectories to stress-test monitors — DanielKhashabi · 2026-10-26
- Hugging Face ships hf CLI updates: deduped cache saves 19GB, faster buckets, retryable GPU jobs — yb2698 · 2026-10-26