Skip to content
AI Agents

Agent vs Agent Negotiation Sandbox

Watch two agents with secret goals negotiate a deal, then try beating one yourself.

Intermediate16h1–3 people16/20 fit
TL;DR

Build Agent vs Agent Negotiation Sandbox for students learning negotiation and ai researchers in ~16h with 1–3 people. Stack: LLM + LangGraph + Next.js.

Why it wins: Interactive and fun with a clear scoring loop.

Build flow

01 · Problem
Negotiation is a key skill but practice partners are rare and lack variety.
02 · Solution
Give two agents private goals and limits, run a turn-based negotiation, score the outcome and let a human take over one side.
03 · Build
  1. 1Define scenarios and private limits
  2. 2Run agents in a LangGraph loop
  3. 3Score deal value
  4. 4Add a human seat
04 · Demo
Run a used-car negotiation and then let a judge negotiate against the agent.

MVP checklist

  • Secret goals
  • Turn-based negotiation
  • Outcome scoring
  • Human takeover

Tech stack

TechRole
LLMLanguage model
LangGraphMulti-agent graph
Next.jsFrontend + API routes

Reference repos

Open-source projects you can build on. Links checked 2026-10-31.

Related ideas

More AI Agents

  • AI Agents

    Multi-Agent Decision Council

    Several agents with different perspectives debate your decision and summarise trade-offs

    Intermediate14h1–2
    LLMLangGraphNext.js
    16/20

  • AI Agents

    Inbox Triage Agent with Approvals

    An agent that labels, drafts replies and archives email but asks before sending

    Intermediate18h1–3
    LLMGmail APILangGraph
    17/20

  • AI Agents

    Research Paper Reading Agent

    The agent walks you through a paper section by section and builds a glossary

    Intermediate16h1–2
    MarkItDownRAGLLM
    17/20