Bot-to-Bot Commerce: AIs Negotiating with AIs
Multiple AI agents coordinating, negotiating, and transacting with each other in a shared environment. The future of automated business operations.
What You'll Build
A multi-agent setup where your AI agents communicate, negotiate, and coordinate with each other. Think of it as building a team of specialists that work together without you playing middleman.
Why This Matters
One AI agent is useful. Multiple agents that can talk to each other? That's a business.
- Purchasing agent finds the best deal and sends it to the finance agent for approval
- Content agent writes a post and the scheduling agent picks the optimal time to publish
- Sales agent qualifies a lead and hands it to the onboarding agent
Each agent is simple. Together, they run operations.
The Concept
Most AI setups are hub-and-spoke: you talk to the AI, the AI does a thing, reports back. Bot-to-bot commerce is peer-to-peer: agents talk to each other, make decisions, and only escalate to you when needed.
Example: Automated Vendor Management
- Inventory Agent notices you're low on supplies
- It messages the Procurement Agent: "Need 500 units of X"
- Procurement Agent emails 3 vendors for quotes
- Quotes come in. Procurement Agent picks the best one (or negotiates)
- It messages the Finance Agent: "Best quote is $2,400 from Vendor B"
- Finance Agent checks the budget, approves, and triggers payment
- You get a notification: "Ordered 500 units of X for $2,400. Delivery Thursday."
You didn't do anything. Six steps happened autonomously.
Prerequisites
- OpenClaw running with multi-session support
- At least 2 agent configurations
- A shared communication channel (Telegram group, internal message bus, etc.)
Step 1: Define Your Agent Roles
Keep each agent focused on one job:
- Agent A: Monitors inventory/triggers
- Agent B: Handles procurement/vendor comms
- Agent C: Manages budget/approvals
Simple roles, clear boundaries.
Step 2: Set Up Communication
Agents need a way to talk. Options:
- Shared Telegram group: Each agent has its own bot token, all in one group
- Internal sessions: OpenClaw's session-to-session messaging
- File-based: Agents read/write to shared files (simple but effective)
Step 3: Define the Protocol
Agents need shared rules:
- How to request something ("REQUEST: need 500 units of X by Friday")
- How to respond ("APPROVED", "DENIED: over budget", "COUNTER: can do 400 units")
- How to escalate ("ESCALATE: need human decision on...")
- When to notify you vs. handle autonomously
Step 4: Set Guardrails
This is critical. Multi-agent systems can amplify mistakes fast.
- Spending limits per agent and per transaction
- Approval thresholds: anything over $X requires human sign-off
- Circuit breakers: if errors spike, everything pauses
- Audit logs: every inter-agent message is logged
Step 5: Start Small, Scale Up
Don't build a 10-agent system on day one. Start with 2 agents doing one workflow. Get that solid. Then add a third. Then a fourth.
Real-World Applications
- E-commerce: Inventory agent + pricing agent + marketing agent
- Content business: Writer agent + editor agent + publisher agent
- Services: Lead qualifier + proposal writer + scheduler
- Operations: Monitor + responder + reporter
Tips
- Keep it observable: You should be able to see every message between agents
- Fail gracefully: If one agent goes down, others should wait, not crash
- Test with mock data first: Let agents negotiate over fake scenarios before real ones
- Version your protocols: As you change how agents communicate, track the changes