Building a buyer agent that negotiates 8 seller agents: what broke and what fixed it
mv_soura · reddit · 2026-10-31
For a hackathon the author built a group-buying system: one negotiator agent bargains against 8 seller agents, each with private floor prices and sealed offers.
Key design: keep money out of the models' hands — agents only set reserve prices and pick targets; a deterministic engine decides outcomes. 200 random auctions ran with zero rule violations.
Pitfalls and fixes (Nemotron 3 Ultra, thinking off for speed):
- Repeated calls to already-answered read tools → remove each tool once answered
- Messaging all sellers at once instead of picking one → prompt fix
- Ending turns with prose instead of tool calls → one fixed reminder message
- Smaller model returning prices as text like "$39.95" → disallow strings in the schema
Guardrails: strict JSON schemas with code-level revalidation, per-negotiation and daily spending caps, labelled replay instead of failure at cap.
Live run: 601 buyers, 5 rounds, 34 seconds, 49 model calls, under 2 cents total.
More from coding & agent
- Creator builds shipwreck diving game with Opus 5.5 and Astrocade MCP — _amirabs · 2026-10-31
- GitHub Universe talk: 'Build once, run on any agent' with slides and skill files — 0xkarasy · 2026-10-31
- Codex Remote live activities go live in the ChatGPT iOS app for tracking agent tasks — SIGKITTEN · 2026-10-31
- Open-source pixel FX bring a 3D wizard battle to life with slimes, archers and a golem — majidmanzarpour · 2026-10-31
- Redditor builds personal agent with Gmail, Docs and Todoist, mulls open-sourcing — Sweaty_Rich7217 · 2026-10-31
- Krea 2 + H3 morph workflow auto-generates soundtracks via SigLIP2 and Suno MCP — Accurate_Public4674 · 2026-10-31