I'm Building the NegoAI Platform in the Open

Dear Negotiation Explorer,

Last week I told you what I watched in Paris: three research teams showing that AI keeps getting better at the very parts of negotiation we call human, and my own answer from the stage, that the way forward is the negotiator with an expert AI on their side.

This week I can tell you what I'm doing about it.

I'm building the NegoAI platform. And I'm going to build it here, in the open, over the next twelve weeks at least.

Readers of this newsletter have met several of my agents one by one. Deepak, the negotiation preparation expert built through the eight-phase framework I published on SSRN. Kahneman, who reads the person across the table. Kraljic, the procurement strategist. Von Neumann, the counterpart simulator you can rehearse against before the real conversation.

There are others I've built and never properly introduced here. Rackham, the sell-side expert, who owns the selling motion from discovery to defending your price. Bruner, the merger and acquisition dealmaker. Sander, the mediator, for when talks deadlock. And the Compiler, the quiet one that folds everyone else's work into a single report.

Each of them works. I use them every week, in my own negotiations and in my teaching. But they live where I built them. To use them together today, you either need me at the keyboard or you need to build your own versions from the write-ups.

That gap is the platform.

Every AI negotiation tool you can buy today is a black box. You type your situation in, advice comes out, and everything in between is sealed: what the tool knows, how it reasons, what happens to the deal information you just typed. You're asked to trust the label.

I've taken the opposite route with my research. Both papers behind my agents are free on SSRN, methodology included, so anyone can check the work. The platform will be built the same way. Every week, this newsletter will carry the build report: what got built, what it cost, what broke, which decisions were made and why. If I skip a week or two in August, that's the summer holidays, and I'll say so. When I get something wrong, you'll read about it here the same week it happens.

I know of no AI negotiation product that was built this way. That alone seems like a good reason to try.

The platform is too big to build all at once, so the first decision of the build is the split.

The first work goes into the app: preparation plus practice, the two things this newsletter has been teaching all year, in one place. You open it and describe the negotiation in front of you: the deal, the counterpart, what you already have. The right experts prepare you in parallel, each through its own lens: the selling or the buying view, the behavioral read of the other side, the negotiation strategy. You get one preparation report. Then, before the real conversation, you practice it. Von Neumann plays your counterpart, pushes back where your strategy is weak, and debriefs you at the end.

After the app comes the full platform, for everybody who carries deals: sellers defending prices and renewals, buyers running sourcing and supplier negotiations, dealmakers in mergers and acquisitions, and managers building international partnerships. A team of expert agents, each with its own lens, covering the work around the whole deal, with negotiation judgment as the spine. Think of it as an operating system for negotiation work: the agents underneath, one platform on top.

The reason it should exist fits in one sentence: today's software automates the routine machinery around deals, the intake forms, the quotes, the records, while the judgment work, the part that decides what the deal is actually worth, gets no expert support at all.

Here's the state of the build as of today, so this series starts truthful.

Nothing is live. There is no app to click yet. The state of the build today is a document: the PRD, the product requirements document, the same compass I described when I told you how Deepak was built. It defines what we're building, for whom, and why, and it's where the split above is written down.

Since the PRD is this week's whole build, let me open it properly.

How it's written, first, because the form matters as much as the content. Every section carries one of three labels: decided, provisional, or open. Decided means a dated decision stands behind it. Provisional means it's the current best thinking, still to be stress-tested. Open means the call hasn't been made yet. A product document that admits what it doesn't know is a compass. One that pretends everything is settled is a wish list with headings.

What's inside: it starts from the people who will use it. The seller whose week is discovery calls, proposals, and defending a price increase. The buyer running a thirteen-step sourcing process in which negotiation is one step of the thirteen. The dealmaker in mid-market mergers and acquisitions who needs partner-grade analysis without partner rates. The manager structuring distributor and partnership deals across borders. For each of them, the PRD maps the deal work they actually do, then which expert agent covers which part of it, and only then what they will click: preparation reports, analysis of the documents a deal produces, rehearsal against Von Neumann, guided runs for the tasks that repeat every week, and a place where a team writes down how it negotiates, so every analysis respects it.

The quality bar is decided. The platform has to hold the level the agents already reached. Last week I wrote that Deepak's preparation goes beyond what the best human negotiators produce on their own. That's the standard for everything that ships.

Among the open: what the platform should cost, and which piece goes live to users first. Those calls will be made during the build, reported here, and replies from readers will carry real weight in them.

Beyond the PRD: the agents you just met, working. Two published papers holding the method. And my partner in the build, Claude Code, which I call Brama. Regular readers have met Brama before.

Two decisions landed this week, and they are this week's report.

The first is the name. I spent time trying out product names, some of them clever, none of them right. Then I noticed the answer was already on the masthead. The platform is called NegoAI. One name for the research, the newsletter, and the product: what I publish, what you read, and what I'm building are the same work.

The second is how it will be built. The building itself will be done by agents. The platform will be built around Hermes, an open-source agent harness, and the first agent standing up on it is the orchestrator: the one that will run the fleet, keep the work moving, and report to me. It doesn't have a name yet. Naming it is next week's decision, and you'll read it here. The whole team you met above moves to Hermes too, agent by agent, and you'll watch that move happen in these reports. Brama stays beside me through all of it.

If you take one thing from this first report, take the PRD move. Before you build anything with AI, an agent or a whole platform, write one page that says what you're building, for whom, and what a good result looks like. It's the same compass at every scale, and it's the single habit that separates builds that converge from builds that wander.

Three things, concretely.

First, you'll watch how one negotiation professional and his agents build a real platform, decision by decision. The method scales down: the same way of working applies when you build a single agent for your own deals, and I'll keep pointing out where.

Second, you get a say while the decisions are still open. A platform built in the open can be shaped by the people it's for. Reply to any issue in this series and you're talking to the builder, before the concrete sets.

Third, every build report will aim to leave you with at least one thing you can use the same week, with or without the platform.

If the platform could take one piece of your deal work off your desk on day one, which piece would you hand it first: the preparation, the reading of your counterpart, the practice before the conversation, or something I haven't thought of?

Just hit reply and tell me. I read and answer every one.