Anthropic preps AI agents to negotiate prices and complaints
Anthropic is asking law firms how AI agents should haggle prices and settle complaints for users. Your customers' agents are coming. Write your floor down.
Anthropic is talking to law firms about how AI agents should negotiate for the people they work for: haggling over prices, settling complaints, pushing for better contract terms. That's per the Financial Times, as reported by The Decoder on October 18. There's no product yet. But if you sell anything, the other side of that negotiation is you.
What actually happened
- The research: Anthropic is working with law firms to set fair negotiation norms for agents that act for users. The tasks named are price haggling, complaint resolution, and better contract offers. The firms are not named.
- The hiring: Anthropic is also hiring "Conceptual Reasoning Fellows," lawyers among them, to spend four months testing Claude on judgment-heavy questions in law, economics, and related fields. Pay is $100,000 to $167,000.
- Why it's hard: Code and math have answers you can check. Law mostly doesn't. Anthropic's Justin Curl told the FT that in law, quality is "notoriously hard to define."
- The earlier evidence: In Anthropic's Project Deal experiment, 69 employees each got $100 and a Claude agent to buy and sell for them. Agents on the stronger model, Opus 4.5, got better prices than agents on Haiku 4.5. The same item sold for $3.64 more on average. People with the weaker agent rated the deals as just as fair. They didn't notice they lost.
Why AI agent negotiation matters for your business
Today a customer who wants a discount sends one polite email and gives up. An agent doesn't get tired and doesn't get embarrassed. It will ask every time, quote your own policy back to you, and try again.
If your discount rules live in someone's head, an agent will find the soft spot. Three things to do now:
- Write the floor down. Lowest price, maximum discount, who can approve more. If a rule isn't written down, a polite and patient agent will get around it.
- Make refund and complaint rules exact. "Case by case" means "whoever asks most." Put it in numbers and days.
- Check your own agent's model. If your support bot negotiates, Project Deal says a cheap model can lose money quietly. Test it against a strong one before you trust it with a margin.
Key takeaways
- Anthropic is consulting law firms on norms for AI agents that negotiate prices, complaints, and contracts
- It's hiring lawyers as fellows at $100,000 to $167,000 to test Claude's judgment
- In Anthropic's Project Deal, stronger-model agents got better prices and the other side didn't notice
- Write down your price floor, discount limits, and refund rules before agents start asking
- If your own bot negotiates, test which model it runs on
Ready for customers who send an agent instead of an email? We build support and quoting systems with hard limits in code, so the margin holds no matter who asks. See what we build or talk to us about your pricing rules.
Sources: The Decoder, citing the Financial Times, Anthropic: Project Deal, The Decoder on Project Deal.
- #anthropic
- #ai-agents
- #agent-negotiation
- #pricing-policy
- #customer-service
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
WhatsApp help pages hid AI instructions on 80 pages
Redact found 102 hidden AI instruction blocks on 80 WhatsApp help pages, one telling the bot not to suggest a regulator. Audit what your AI agent reads.
Read itTenstorrent raises $1.5B at $6.2B for open-weight inference
Tenstorrent raised $1.5B at a $6.2B valuation to build RISC-V inference servers for open-weight models. What it means for your AI hosting costs.
Read it