Is the demo making a real decision?

No. Everything here is scripted so it plays the same way every time. It shows how an adversarial agent system reasons, not a live model making a live call. In a real build the agents run for real, on your data, with guardrails and a human in the final seat.

Are the figures real?

No. They are illustrative and clearly labelled. We made them plausible so the logic is recognisable, but they describe no real deal, candidate or company.

Why four agents arguing instead of one answer?

Because a single answer is easy to dismiss as autocomplete. Forcing AI to argue against itself surfaces the conditions and risks a one-shot reply glosses over, and leaves you with a conclusion you can defend to a board, a lender or a team.

Could this run on our own decisions?

Yes. The same pattern applies to underwriting, hiring, supplier choices and any other high-stakes call. We ground the agents in your data, set the guardrails, and keep a person making the final decision. That is the build behind the demo.

Adversarial agents

Watch four AI agents argue a decision, then agree on what to do

Most AI on the web is a chatbot answering a question. This is AI reasoning against itself and reaching a call it can defend. Press one button and watch it happen.

Commercial real estateIllustrative data

Multi-let office, regional city centre

An off-market acquisition put to the investment committee.

Asking price: £8.2m
Net initial yield: 7.4%
WAULT: 4.2 years
Occupancy: 86% (3 of 14 suites void)
EPC: Average D, two units E
Debt offered: 60% LTV at 6.1%

The decision on the table

Buy at the asking price, chip the price, or walk away?

Investment Committee

4 membersGroup chat

Press Run the debate to start the conversation.

Four agents join the chat. Watch them argue it out in real time.

Illustrative and scripted. The agents do not call a live model here, the figures are invented, and this describes no real deal, candidate or company. In a real build the agents run on your own data, with a person making the final decision.

How it works

Four agents, two rounds, one defensible call

This is not a chatbot answering a question. It is four AI agents, each given a different job and a different incentive, arguing the same decision until they reach a conclusion they can all stand behind.

Step 01
Four roles, four incentives
A champion makes the case to act, a sceptic stress-tests it, a money role guards the numbers, and a stakeholder speaks for the people the decision lands on. Each is briefed to push its own view as hard as it can.
Step 02
Two rounds of argument
In the first round the agents stake out positions. In the second they respond to each other, which is where the weak arguments fall away and the real conditions surface.
Step 03
A resolved verdict
A final pass weighs the exchange into a clear recommendation, the conditions attached to it, and two risk-adjusted ways to act on it.

No overclaiming

In a real build, these agents are grounded in your own data, not invented figures, and they draft the decision rather than make it. A person always makes the final call and signs it off. This page is scripted to show the shape of that reasoning, and every figure on it is illustrative.

The honest bit

This pattern applies to your underwriting, your hiring shortlists, your supplier decisions, any call where being able to defend the answer matters as much as the answer itself.

See what this could do for your decisions

Watch four AI agents argue a decision, then agree on what to do

Multi-let office, regional city centre

Four agents, two rounds, one defensible call

Four roles, four incentives

Two rounds of argument

A resolved verdict

The honest questions about a demo like this

See the pattern at work

Property AI Lab

Fractional AI leadership