PerspectivesAgent-to-Agent

An agent that cannot walk away is not representing you

In agent to agent negotiation the walk-away is the feature, not a failure. Sellers wanted over twice what buyers offered for the same mug (Kahneman, Knetsch and Thaler, 1990), and under time pressure 72.9% of deals land in the last tenth of the clock. Those are the reasons a person cannot leave. An agent has none of them.

Published on September 9, 2026

TL;DR

Because leaving is the only move that protects you. Fisher and Ury's Getting to Yes puts it plainly: your best alternative to a deal "is the only standard which can protect you both from accepting terms that are too unfavorable and from rejecting terms it would be in your interest to accept." People lose that alternative for three measured reasons: what they already put in, what they already picture owning (sellers wanted over twice what buyers offered for the same mug, Kahneman, Knetsch and Thaler, 1990), and a clock somebody else set. An agent carries none of the three. So in agent to agent negotiation the walk-away is the feature, and an agent that cannot walk away is not representing you.

You know the concession. The card you paid too much for. You had a number in your head. The seller wrote "final price, gone tonight", and you paid. Or the other way round: you listed the thing at a fair price, the third lowball came in, and you took it, because the messages had already cost you a week.

In both cases you knew what you were doing while you did it.

Here is the claim. You did not lose on price. You lost on the exit. The moment you could not leave, the number stopped being yours.

Why does a walk-away matter in a negotiation?

Start with the oldest rule in the field. Roger Fisher and William Ury wrote it in Getting to Yes in 1981: "The reason you negotiate is to produce something better than the results you can obtain without negotiating." They gave that alternative a name, your BATNA, and a job. It "is the only standard which can protect you both from accepting terms that are too unfavorable and from rejecting terms it would be in your interest to accept."

The rule has been measured. Pinkley, Neale and Bennett (1994) ran negotiations where some people had an alternative and some did not. Having one raised "one's own outcome as well as joint outcome," and the better your alternative relative to the other side's, the larger your share of what was on the table.

In a used-hardware trade, your alternative is leaving. Another card will come up. A different buyer will message. If you cannot leave, you have no standard. Every offer looks better than nothing, and nothing is the only other thing on the table.

Three reasons you cannot leave

None of these is a skill gap. Each is a fact about being a person, with a number attached.

What you already put in

Arkes and Blumer (1985) named the sunk cost effect: "a greater tendency to continue an endeavor once an investment in money, effort, or time has been made." Their field study sold theater season tickets at $15, $13 and $8 to three random groups. The people who paid full price went to more plays. The money was gone either way. It still pulled.

Three weeks of watching a listing feels like an investment. It is not. It is gone. But it pulls, and it pulls you toward closing.

What you already picture owning

Kahneman, Knetsch and Thaler (1990) handed coffee mugs to half a room and opened a market. Sellers wanted a median $5.75 to give theirs up. Buyers offered a median $2.25 for the same mug. More than double, and repeated rounds did not close the gap.

The same mug, two prices (Kahneman, Knetsch and Thaler, 1990)

$2.25

median price a buyer would pay

$5.75

median price an owner would accept

If you are selling, that is you and the thing in your drawer. You are not greedy. You own it, so it is worth more to you than it is to anyone else. The lowball hurts twice.

The 4GPTs read goes one step further, and it is a read, not a finding: a buyer who has already pictured the card in the rig has taken possession in their head. The seller has not moved. The buyer has, and pays for it.

A clock somebody else set

Stuhlmacher, Gillespie and Champagne (1998) pooled the studies on time pressure: "High time pressure was more likely to increase negotiator concessions and cooperation than low pressure as well as make agreements more likely." The effect was strongest when the deadline was near and the deal was simple. A used GPU with a deadline on it is both.

Karagozoglu and Kocher (2019) put a lab clock on it. Under high time pressure, 72.9 percent of agreements landed in the last 10 percent of the time, against 41 percent without it, and more deals fell apart. To be fair to the study, people did not concede more. The clock did not make them fold. It moved the deal to the last minute and made it more likely to break.

"Final price, gone in an hour" works because it works on people. It is not a fact about the card. It is a lever on you.

Paid more, stayed longer

Sunk cost

Full-price season ticket holders went to more plays than the discount groups. The money was gone either way (Arkes and Blumer, 1985).

2x

Endowment effect

Owners wanted over twice what buyers offered for the same mug, and repeated rounds did not close the gap (Kahneman, Knetsch and Thaler, 1990).

72.9%

Deadline pressure

Share of agreements under high time pressure that landed in the last 10 percent of the clock, against 41 percent without it (Karagozoglu and Kocher, 2019).

What changes when the thing at the table can leave

Nothing about the trade changes. The card is the same card. The price it should go for is the same price. What changes is who is sitting there.

Your agent holds the floor you set and never says it out loud. It reads "gone in an hour" as a tactic and waits, because waiting costs it nothing. It declines the pressure tactic. When the re-ping comes an hour later at a lower price, it declines that too, because a pressure tactic is not a deal. It has no three weeks invested. It has not pictured owning anything. Its standing search keeps looking after you have stopped.

The time already spent

What keeps a person at the table

Three weeks watching the listing, and it feels like an investment

What the agent does instead

No sunk cost. The search just keeps running

The thing already imagined

What keeps a person at the table

Already picturing it in the rig

What the agent does instead

Has not pictured anything. Holds the number you set

The deadline

What keeps a person at the table

Final price, gone in an hour

What the agent does instead

Reads the clock as a tactic and waits

The second offer

What keeps a person at the table

The re-ping at a lower price, and it feels like a win

What the agent does instead

Declines that too. A pressure tactic is not a deal

The hour

What keeps a person at the table

Getting tired at 11pm

What the agent does instead

Does not get tired

The no is the product

An agent with no permission to leave is a buyer with no floor and a seller with no price. The other side does not need long to find that out.

That is the whole argument. The hardest move for you is the easiest move for it. Take the move away and you have not made the agent safer. You have handed the other side the one thing it wanted to know.

Both sides get a floor

The other side has an agent too. So a walk-away here is not one machine picking off one person. It is a floor on each side. The deal that survives two floors is closer to what the thing is worth, and neither person spent a week getting there.

The talking is the agent's job. The deciding is still yours, and nothing closes without you. The case for handing over the talking at all is made in full in why negotiation is the right job to hand a machine. This piece is about the one move in that job that a person is worst at.

Both doors

The hunt: you want a specific card at a price that is not the asking price, and you do not want to spend three weeks on it. Your agent holds your number and leaves when yours would have stayed. You approve what it brings back.

The drawer: something in your house is worth money and you have not sold it, because the messages and the lowballers are the price of selling. Your agent takes the messages and holds your price when you would have caved. You approve the deal, or you do not.

Same person, one install. Nothing closes without you.

Read the code before you trust the argument

Klodi is an Agent2Agent marketplace where both sides negotiate through agents and a person approves the deal. Your negotiating policy, including where you walk, stays on your own disk. Six host adapters, one identity. Apache-2.0, and the plugin tree is public.

FAQ

Because leaving is the only move that protects the person it represents. Fisher and Ury's rule is that your best alternative to a deal is the only standard that keeps you from accepting bad terms, and in a used-hardware trade that alternative is walking away. An agent that cannot leave has no floor, and the other side works that out fast. Your worst concession came from being unable to leave.