The deal you never made

For decades, negotiation was rationed by human time. AI Agents are ending that rationing.

Hi, and welcome back to the Situation Room.

Each week we track what is happening at the frontier of Agentic Commerce and translate it into something you can actually use. Not the hype, not the vendor noise, but the shifts that will change how your customers find you, how your business operates, and how entire industries organise themselves around AI agents acting on behalf of people and organisations.

This week we are looking at negotiation. Specifically, why most businesses are leaving serious value on the table every day, why that has been unavoidable until now, and why agents change the calculation entirely. This one has implications that go well beyond procurement.

Enjoy the read.

In 20 seconds

Scheduling a meeting across three organisations used to require a phone call, a competent assistant, and a bit of goodwill. Now it requires three Calendar links, four rounds of email, and a meeting that happens next week instead of this one. The technology made it worse, not better, because the technology removed the conversation. And it was the conversation that would have revealed that the person who "cannot do Tuesday" can actually do Tuesday, if someone asks the right question.

That is the thing about constraints. Many of them are not real. They just look real until someone takes the time to probe them.

Business is negotiation

In Spanish, the word for business is negocio. The word for negotiate is negociar. They come from the same root, and in everyday Spanish the connection is obvious in a way that English obscures. Business and negotiation are the same activity, described from slightly different angles. Every commercial relationship, at its origin, was a conversation about terms.

Scale made that impractical. A supermarket cannot negotiate the price of every packet of pasta. A procurement team cannot run a sourcing event for every five-thousand-euro purchase. So businesses settled for posted prices, rate books and standard contracts. That was a reasonable trade. It also means that most businesses now run on terms that were set once, accepted as fixed, and never questioned again.

The gap between the terms you have and the terms you could have, across an entire operation, is larger than most finance teams realise. BCG estimates that actively managing tail spend alone, the 80 percent of transactions that fall below procurement engagement thresholds, can reduce annual expenditure by 5 to 10 percent. For a business with 500 million euros in tail spend, that is up to 50 million euros. Not because the deals are bad. Because no human has the bandwidth to renegotiate them.

What agents bring that humans cannot

This is where the conversation about agents usually goes wrong. People frame agents as faster workers. Cheaper automation. A way to process more tickets or summarise more documents. That framing misses what makes agents genuinely different for negotiation.

An agent is not emotional. It will not lose patience on the hundredth supplier conversation of the day. It will not skip preparation because it is tired or pressed for time. It can hold a wide perspective across complex dependencies, monitor changes across a supply network, microtask across thousands of individual transactions, and still apply broad knowledge to each one as if it were the only negotiation happening. Agents bring exactly the traits that make negotiation scale: patience, diligence, breadth of context, and no ego investment in being right.

A human expert negotiator is a scarce, expensive resource deployed selectively. An agent negotiator is available for every transaction, every renewal, every routing decision, every service level conversation that currently gets waved through because nobody has the time to look at it.

Most systems are under-negotiated, not under-resourced

The deeper argument is this: most businesses and supply chains run sub-optimally not because the people running them are incompetent, but because it is genuinely impossible for any human team to take a bird's eye view of an entire ecosystem and make the thousands of small adjustments that would optimise the whole. Buffer stock exists because parties do not trust each other to have the right information when supply is disrupted. Contracts auto-renew because nobody has time to read them. Logistics routes go unquestioned for years because renegotiating them falls below the threshold of what is worth a buyer's week.

When the World Food Programme's supply chain director described what coordination failure actually looks like at scale, she was not describing negligence. She was describing a system under more pressure than any human team could manage manually. When Middle East conflict disrupted Red Sea shipping routes, food for Afghanistan sourced in Pakistan had to be rerouted through Iran, diverted to Dubai, then trucked through Saudi Arabia, Jordan, Syria, Turkey, Azerbaijan and Turkmenistan to reach Kabul, adding 1,000 euros per ton and three weeks per delivery. WFP's team negotiated waivers, rerouted cargo and managed carrier relationships under extreme time pressure, and they did well. But they were doing as a crisis response what a well-designed multi-agent system could be doing continuously, before the crisis, as normal operations.

There is no shortage of food in the world. It is in the wrong place at the wrong time, because the right negotiations have not happened. The same is true of most supply chains, and most procurement operations, and most service contracts. The constraints are not fixed. They have just never been examined, because examining them at scale has never been affordable.

William Ury, co-author of Getting to Yes and co-founder of the Harvard Program on Negotiation, has spent decades studying what happens when parties actually surface their constraints rather than assume them. His observation is straightforward: in the vast majority of negotiations, the things each party thinks are immovable turn out not to be, once someone takes the time to ask. The deal that seemed impossible becomes possible because one party needed something the other did not value, and neither knew it.

That surfacing takes time. Skilled negotiators spend hours in preparation before a single offer is made. They model the other party's priorities, identify what they value least that the other party values most, and look for the trade that benefits both. It is good work. It is also expensive, which is why businesses reserve it for transactions where the margin justifies the cost.

Agents make it affordable everywhere. They can be instructed to find the win-win, given the parameters within which to find it, and trusted to run the process without cutting corners because they are tired or because the deal is small. They will do the preparation every time. They will ask the question about Tuesday.

Do you want Situation Room updates delivered to your inbox?

What good negotiation looks like at machine scale

The research on how to design agents that negotiate well rather than extractively is accumulating fast. Researcher Qi Han Wong ran a controlled experiment pairing Claude and Gemini as buyer and seller across three negotiating styles. The polite agent, negotiating with relationship-first language, reached a better deal in fewer rounds than the transactional one. The aggressive anchor produced a deadlock and left value on the table for both parties.

​MIT's analysis of 180,000 AI-versus-AI negotiations confirmed the pattern at scale: agents that prepared before negotiating, maintained a warm tone throughout, and built firm boundaries around their own private information consistently outperformed agents that optimised only for extraction. The finding that matters for business leaders is not the methodology. It is that agents instructed to create value close more deals than agents instructed to claim it.

Three companies already building on this

​Causa Prima​ is building an agentic network for cross-company finance operations, founded by alumni of Taulia, Centrifuge and Oak Security. Early product, mature architecture, worth watching as the cross-company negotiation layer develops.

​Pactum negotiates thousands of supplier contracts in parallel for Walmart and Maersk, the clearest proof that autonomous negotiation at industrial scale is already operational, not experimental.

​Procure.ai automates tail spend negotiation end to end, with published customer data showing 4.9 percent average saving per event across procurement that previously went untouched.

The governance layer most organisations have not yet built

Building agents that negotiate well also requires infrastructure that most organisations have not yet put in place. The AP2 protocol addresses part of this, providing cryptographically signed mandates that create an auditable chain from intent to agreed terms to payment. What it does not yet standardise is the deal formation that precedes the transaction: how offers are exchanged, how concessions are structured, and what record is produced when negotiations end. Without standard outcome codes and signed records, the output of an agent negotiation is a conversation log. Conversation logs do not hold up in commercial disputes. The organisations that build this infrastructure early will delegate negotiation safely. Those that skip it will find out why it matters after the first significant dispute.

I discuss this more in the post Where Agents Make the Deal​

Trusted Agents

We track how this is developing across sectors so you have a clear picture before you need to make a decision. If this edition has raised questions specific to your organisation, where your negotiation gaps are, what parameters you would need to set before delegating safely, or how to read the protocols and platforms emerging in this space, we offer customised briefings built around your sector and your roadmap. Consider us your outsourced product radar for agentic commerce.

One question to close with

Where in your business does a deal not happen, not because it is a bad deal, but because no human has the time to have the conversation?

Until next week.

Gam

​The Trusted Agents Ltd.​
​Unsubscribe · Preferences​