The most underrated fact about modern advertising is that it already runs on machine negotiation. Every programmatic impression is an auction between algorithms, resolved in about a tenth of a second, and it has worked that way for over a decade while most marketers treated it as plumbing. I explained the mechanics years ago in programmatic advertising explained, and the reason I keep pointing back to it is that programmatic was a preview. It showed what happens when both sides of a market become software: the market speeds up, granularizes, and re prices continuously.
Now the same transformation is coming for the demand side of commerce itself. Buying agents will not just select products, they will negotiate for them: price, bundles, delivery terms, warranty extensions, data permissions, even the terms under which they will accept promotional messaging at all. On the other side, merchant systems will negotiate back, in real time, per transaction. Two pieces of software haggling is not science fiction. It is a pricing API meeting a purchasing agent, and the primitive versions already exist.
My claim in this paper: by 2030, fixed public pricing will be a minority position in agent heavy categories, replaced by negotiated ranges resolved per transaction, and the advertising industry's auction infrastructure is the template for how it will work. The strategic question for brands is not whether to negotiate with machines but at which layers, with what guardrails, and how to avoid being systematically out negotiated by better tooling.
Key Findings
- Machine to machine negotiation collapses the cost of haggling to near zero, which means everything becomes negotiable, including things that were never negotiable at retail.
- Negotiation will stratify into distinct layers: price, terms, bundle, data, and attention. Each layer has different economics and different risks.
- The side with better information wins repeated negotiations, and merchants start with a data advantage they will squander if they respond with crude discounting.
- Buying agents will learn merchant concession patterns at scale, so any exploitable discount logic will be found and drained systematically.
- Advertising budgets partially convert into negotiation concessions: the discount an agent extracts is functionally a performance marketing cost.
- Categories with high margin variance and perishable inventory flip to negotiated pricing first: travel, events, fashion clearance, services.
Why Haggling Comes Back
Fixed pricing is historically weird. For most of commercial history, price was a conversation. Fixed tags won because negotiation does not scale: you cannot haggle with ten thousand daily customers across a counter. The department store did not end negotiation because customers hated it. It ended it because labor economics made it impossible.
Software removes that constraint completely. An agent negotiates at zero marginal cost, in parallel, forever. Once buyers carry tireless negotiators, sellers who refuse to negotiate are simply leaving a lever unpulled while competitors pull it. Game theory does the rest. This is why I am confident about direction even while humble about timing: the equilibrium of cheap negotiation is negotiation everywhere it improves outcomes for either side.
What returns will not look like a bazaar. It will look like structured concession spaces: merchants exposing machine readable rules about what can flex, price within a band, shipping upgrades, bundle discounts, loyalty terms, and agents probing those spaces for the best package. Most humans will never see it happening. They will just notice that their assistant "found a better deal," the way they once noticed a good coupon.
The Five Layers
Treating negotiation as one thing is the mistake that will cost merchants the most. There are at least five distinct layers, and conceding at the wrong layer is how margin dies.
The price layer is the obvious one and the most dangerous, because price concessions are permanent education. Every discount an agent extracts becomes training data about your floor.
The terms layer, delivery speed, returns windows, warranty length, payment timing, is where smart merchants will steer negotiations, because terms have asymmetric value: a flexible delivery date might cost you nothing operationally while being worth a real concession to the buyer's mandate.
The bundle layer is the merchant's best friend. Bundling obscures unit price, moves the negotiation to package value, and lets you concede visibly while protecting the number that anchors your category position.
The data layer is new and underpriced. Agents can offer something merchants desperately want: verified purchase intent, mandate context, permission for follow up. Suppose an agent discloses that its principal replenishes monthly, in exchange for a locked price. That disclosure is worth more than the discount, and almost nobody is building to trade for it yet.
The attention layer is the strangest and the one advertisers should watch. Agents will negotiate over whether they will even consider sponsored options, under what labeling, and at what compensating value. When consideration itself becomes a negotiable unit, advertising has formally merged with commerce. I flagged in Paper No.1 that platforms will pilot agent native ad products quietly; this layer is where those pilots will live.
The Concession Learning Problem
Here is the mechanism I would tattoo on every pricing team's wall. Human hagglers are inconsistent and forgetful. Agent hagglers are neither. Every negotiation an agent runs against your systems is a data point in a model of your concession behavior, and agents in a given ecosystem can effectively pool that learning across millions of merchants and transactions.
I call the resulting dynamic Concession Drain: the systematic discovery and exhaustion of a merchant's every exploitable pricing rule by learning agents, at machine speed, without the natural rate limits that human deal seeking imposed. If your system gives a discount to carts abandoned for 48 hours, agents will learn to abandon carts. If your floor moves at month end, agents will learn your calendar. Whatever pattern exists will be found, because probing is free.
The defense is not secrecy, which merely slows discovery. The defense is designing concession logic you are content to have fully known: rules that trade concessions for things of real value, verified commitment, larger baskets, data, timing flexibility, rather than rules that leak margin to whoever asks correctly. Assume perfect surveillance of your pricing behavior and design accordingly. Merchants who bolt agent negotiation onto legacy promo logic will experience it as a sophisticated coupon exploit that never sleeps.
What This Does to Advertising Budgets
Follow the money and the industry implications get uncomfortable. Today a brand spends to acquire a customer, then sells at list price. In a negotiated economy, part of that acquisition spend converts into concessions granted at the point of machine negotiation. The discount an agent extracts is functionally indistinguishable from a performance marketing cost: it is money surrendered to win a specific transaction.
That means media budgets and pricing policy stop being separable departments. The CFO question of 2029 will be: what is our blended cost of winning an agent mediated sale, across ad spend, negotiated concessions, and attention layer payments, and how does it compare to our old CAC? Brands that cannot compute that number will be optimizing two halves of one system independently, which always ends with both halves losing. This is also where owning your customer relationship pays compounding dividends: the more demand arrives through mandates that name you, the less you surrender in open negotiation, a dynamic that makes first party data strategy a pricing weapon and not just a media hedge.
Timeline and Order of Battle
My falsifiable predictions, on the record. Through 2027: negotiation stays mostly implicit, agents exploiting existing coupons, price matching policies, and cart mechanics, while a first wave of merchants exposes deliberate offer APIs. By 2028: at least one major commerce platform ships a formal negotiation protocol for agents, with structured concession types, and travel and services adopt it fastest. By 2030: negotiated ranges are the norm in high margin variance categories, and "list price" functions mainly as an anchor for the band. If 2030 arrives with fixed pricing still dominant in travel and event inventory despite heavy agent traffic, my model is wrong. I do not expect to owe that correction.
What I Would Do About It
Map your concession space now, on paper, before any protocol forces you to. For every product line, decide what can flex, by how much, in exchange for what. The merchants who enter machine negotiation with a designed concession space will run circles around the ones who enter with a panicked discount matrix.
Steer every negotiation away from price and toward terms, bundles, and data. Build your offer logic so the easiest concessions for an agent to win are the ones that cost you least and teach the least about your floor. Price is the concession of last resort and should be priced like one.
Audit your existing promo logic as an adversary. Assume a tireless agent is probing every coupon rule, cart trick, and regional price gap you have. Whatever it would find, fix or formalize. An exploit you convert into an official, value traded offer is strategy. An exploit you leave implicit is Concession Drain.
Start measuring blended win cost. Wire your reporting so ad spend and granted concessions land in one view per transaction cohort. Even a crude version of this metric in 2026 puts you years ahead of competitors who will meet it for the first time in a board meeting they do not enjoy.