|
AI can take four seats at your negotiation table. Each does a different job, and each fails differently.
Last spring, on a video call with a client's deal team, I watched a senior executive answer every question three seconds late.
Every question. The same delay.
His connection was fine. He had a second window open, and he was reading from it. The counterpart noticed before I did; not the window, the rhythm.
He had let AI sit beside him. He had never decided to. That is the part worth your attention.
According to Jared R. Curhan and Jonathan Gratch's research, AI can occupy four positions in a negotiation, and the position matters more than the technology. Each chair has a job, a payoff, and a specific way of going wrong.
- In front: it speaks for you.
- Besides: it advises you.
- Behind: it organizes your own side.
- Between: it runs the room for both sides.
CHAIR ONE · IN FRONT
The AI is your agent
What it is. The system negotiates in your place. It makes the offers, answers the counters, sets the tone. You are not in the conversation. The word for this is delegation, and it is a bigger step than it looks: the AI is not helping the negotiator; the negotiator is.
What it is for. Volume and consistency. A company renewing four hundred freight contracts, a platform settling routine claims, a procurement team working through standard supplier terms, situations where the deal is repetitive, the relationship is thin, and human attention is the bottleneck.
Where it goes wrong. Two design choices sit inside every AI negotiator, and someone you have never met already made both.
The first is whether it is built to calculate or to relate. A system tuned to maximize a score will happily trample the norms that make an agreement feel legitimate, and agreements that feel illegitimate come back to you, renegotiated, resented, or quietly sabotaged. A system tuned for smooth language may charm the room and fail to protect your interests.
The second is whether it is built to compete or to cooperate, to grab the biggest slice, or to grow the pie. That decision lives in a settings file, and it has just become your negotiating philosophy.
Then comes the question no vendor will answer for you. If your agent concedes eight points on payment terms, who conceded? You? The people who built it? The company that bought the license? Delegation does not just move the work — it moves the authorship, and accountability travels with it.
Use it this way. Delegate the task, never the judgment. Set the objectives, write down the trade-offs you will accept, and keep an explicit right to override. Never seat an AI in front of you when the relationship is the real asset. And disclose it: if the other side would feel misled to learn they were talking to software, you have not found an efficiency — you have found a deception.
CHAIR TWO · BESIDE
AI as your coach
What it is. You stay at the table; the AI advises. It works in three tenses. Before, it helps you prepare and rehearse. During, it feeds you live suggestions: what to say, what to ask, when you are talking too much. After, it reads the transcript and tells you how you did.
What it is for. Two things, and the second is the important one.
The obvious one is bandwidth: more preparation, faster, than you would have done alone. The deeper one is access. Real negotiation training has always been a luxury good: small seminars, expensive coaches, personal feedback for people whose employers can afford it. A coach that never sleeps changes who gets to practice, and it matters most for the people who were never invited into that room.
Where it goes wrong. In the middle tense. Live coaching is the most seductive use and the most costly to your judgment.
It fails in two ways. Mimicry: you deliver the suggested line without understanding why it works, and the moment the counterpart goes off-script, you have nothing. Dependency: the prompt becomes the shape of your thinking, and you slowly lose the ability to reason under pressure without one.
The right comparison is driver assistance. The system can expand what you are capable of. It must never displace who is responsible for the car.
Use it this way. Load the ends, keep the middle clean.
- Before the meeting, have it play your counterpart at their most difficult, and tell it to attack your position rather than admire it.
- Afterward, give it the transcript and ask three questions: where did I concede first, which question did I never ask, and what did I assume without testing?
- During the meeting, close the window, or restrict it to retrieving facts, never supplying language. Above all, ask it to explain, not to script. If you cannot say why the advice is right, you have not learned anything. You have only performed it.
CHAIR THREE · BEHIND
AI at your backtable
What it is. Almost nobody negotiates alone. You represent a company, a board, a union, a family. Behind you stands your own side, and this chair belongs to the AI that keeps it coherent: gathering what each stakeholder wants, comparing their priorities, modelling trade-offs across the issues, and finding the contradictions in your own position before the other side does.
This is not coaching. Coaching improves the negotiator. Backtable support manages everything standing behind the negotiator.
What it is for. Internal alignment, which, in serious negotiations, is often harder than the external deal. Frontline humanitarian negotiators use these systems to analyze context, generate options, and weigh risks before they ever enter the room. In mergers, in union contracts, in multiparty environmental agreements, the deal usually dies at home, not across the table.
I have said before that the hardest negotiation is the one behind your own table. Six people who each define "a good deal" differently. A finance lead who has never said out loud that the timeline is fiction. A regional head who will torpedo the agreement in week three because nobody asked her.
An AI is very good at that job, and this is the chair I recommend most often. It surfaces the contradiction your team is politely refusing to name, before that contradiction walks into the room with you.
Where it goes wrong. It flattens. Aggregating everyone into one recommendation turns the lone dissenter into a rounding error, and the dissenter is often right. And it optimizes what it can count: reputation, trust, precedent, and twenty years of supplier goodwill do not appear in the model, so the model will trade them away for nothing.
Use it this way. Treat it as advisory infrastructure, not decision authority. Ask it for your side’s internal contradictions before you ask it for a recommendation. Ask it to build three packages of equivalent value to you and different value to them, so you arrive with real options rather than one number and a hope. Then ask what it could not score and put that back on the table by hand. The Negotiation Canvas® spends its opening lenses on interests, yours and your own side’s and theirs, before a single number appears. That is invisible work; most teams skip it, and this is the chair that can help you do it.
CHAIR FOUR · BETWEEN
AI as mediator
What it is. The system represents nobody. It runs the process: structuring the exchange, summarizing each side’s position, proposing trades that might satisfy both, noticing when the temperature rises, steering toward common ground.
What it is for. Scale and patience. A human mediator can hold one room; a system can hold ten thousand, without getting bored, tired, or partial. It already appears in landlord–tenant disputes, insurance claims, online marketplaces, and in deliberation experiments where structured AI facilitation has helped genuinely polarized groups find agreements they could both sign.
Where it goes wrong. Mediation is not a computation.
A fair process is not one that scores well. It is one where people believe their voice counted. That belief is what makes an agreement stick, and it cannot be manufactured by a summary. If the parties cannot see how the middle is reaching its judgments, they will assume the middle has an owner. They will be half right, because someone wrote it.
Use it this way. Give the AI the labor and keep the judgment human. Let it draft options, summarize positions, and flag escalation; let a person decide how the process runs and what fairness means in this room. And if you are ever a party to an AI-facilitated process, ask the one question that matters: who defined "fair" here, and may I see the definition?
ONE MORE THING
The chairs do not stay put
In practice these roles blur. A tool that prepares you starts drafting your language. A summarizer between the parties starts proposing terms. A backtable system starts whispering in real time.
No system announces the moment it changes chairs. The drift is silent, and the tool you invited in to help you think ends up doing the thinking.
IN PRACTICE
Draw the seating chart
- Name the chair. Write down which of the four the AI occupies in this negotiation. If you cannot name it, it is beside you, and you did not choose that.
- Write the override rule. One sentence, in advance: here is what this system may never decide on my behalf.
- Settle disclosure. If the other side would feel misled learning how you used it, you already have your answer.
- Ask what it cannot measure. Put relationships, reputation, precedent, and dignity back on the table by hand.
- Audit afterward. One question in the debrief: what did the AI decide that I should have decided?
The AI is getting a seat. That was settled the moment your team started opening a second window.
The only thing still open is which chair.
Choose it deliberately, or the software will choose for you, and it will always choose the one beside you.
|