Dear Negotiation Explorer,
Last week, my course cohort built a working negotiation agent in minutes. This week, they used it to do something negotiators have been told for decades they can't: create value without trusting the other side.
That needs unpacking, because it overturns one of the oldest rules at the table.
The rule: to create value, you have to trust.
There are two things you can do in any negotiation. You can claim value — fight over your share of a fixed pie. Or you can create value — expand the pie first, so there's more for both sides to divide. Negotiators call the second one integrative bargaining, and it's where the real gains are.
But creating value has always carried a price: trust. To find the trades that make both sides better off, you have to reveal what you care about most, and what you'd give up. The moment you do, the other side can use it against you.
So most negotiators play it safe. They treat every issue as a fight over one number and walk away with less than the deal could have held. That's the trap: the most valuable move is also the most exposed one.
The finding: AI breaks the trap.
In a study I ran with 120 senior executives on complex deals — published on SSRN as When AI Joins the Table — joint gains rose 84% when both sides had AI support. The part that matters: those gains held even when the two sides didn't trust each other.
That's the headline of my research: AI enables integrative negotiations even in low-trust settings. You no longer need to trust the person across the table to expand the deal. Each side's AI does the integrative work privately — sorting the issues, weighing priorities, finding the trades — so the pie grows without anyone having to open up to get there.
What it looks like in practice.
This week the cohort moved to a new case: the Law Library. One firm is selling a 300-volume collection to another. On the surface it's a price fight — one number, your gain is their loss.
Look closer and there are three issues, not one: price, shipping, and timing. And on timing, the two sides want opposite things. The seller is relocating and needs the books gone within two weeks. The buyer has nowhere to put them for five — every early week costs them storage.
There's the trade. The books leave the seller's library now and ship to the buyer later. The seller hits the deadline; the buyer skips the storage bill. Same deal, both sides better off — and neither had to trust the other to find it. That's value creation, and it's exactly the move an agent surfaces when a person would never risk fishing for it across a tense table.
How we built the agent.
Last week's build was fast because the tool made the decisions — a solid start, but its choices, not yours. This week the cohort built the rest, deliberately, with a method that runs in five steps:
Discovery — understand the problem before writing a line.
PRD — a blueprint of what the agent should do, before how.
System instructions — turn that blueprint into the rules it follows.
Self-evaluation — check those rules against the blueprint, as a skeptic.
Real-world test — run a live case, find what's missing, tighten, repeat.
One discipline runs through all five, and it's the one that decides whether an agent is any good: keep what the agent knows — the negotiation knowledge base — separate from how it behaves — its instructions. Mix them and the agent recites rules it should be reasoning with. Keep them apart and it reasons across what it knows, the way an expert does.
That's where a second kind of trust comes in. You don't have to trust the other side anymore — but you do have to trust your own agent to find the trades it's there to find. That trust isn't free. It comes from building the thing deliberately and testing it against a real case until the output holds up. It's the work last week's quick build skips, and it's the work this week is about.
What this means for you.
The hard part of creating value was never the thinking. It was the exposure — and AI removes it. If you've held back from expanding a deal because opening up felt too risky, that calculation has changed.
You can feel the difference on a deal you're preparing right now. List every issue on the table, not just the obvious one — price, yes, but also timing, volume, terms, risk. Then ask where your priorities and theirs differ. Those gaps are where the trades live, and they're exactly what a well-built agent is there to find for you.
If you'd like a head start, my free tool — The Augmented Negotiator's Brief — turns any upcoming negotiation into a one-page strategic brief, ready to drop into ChatGPT, Claude, or Copilot. Get it free at negoai.ai/subscribe.
The next cohort of the course runs in September. I'll open the details here first — so if building your own agent alongside a group appeals, this is where you'll hear it.
This week's question.
Think about your last hard negotiation. If your AI had quietly found one trade across two issues — without either side having to show their hand — would the deal have been bigger? Reply and tell me. I read every response.