What wargames really show us

The most valuable output of a wargame is a clear view of how people make decisions. AI helps us scale and test that — but whether people trust it is the real challenge.

As Christopher Goldston and I have always said: wargames let us experiment with ideas of change in a safe-to-fail environment. Better to spend a few hours on hypotheticals in the classroom than to find out in the field, when a soldier’s weapon fails to fire or their drone goes down to a cyber attack nobody saw coming. When designing tools like Battle Chess, the goal was never just to show how capabilities layer and interact, or how people respond when the dilemmas start stacking up. It was to build the kind of strategic thinking that can underpin future analysis.

And that thinking always comes back to people. In my experience, the most valuable thing a wargame produces is a clear view of how people actually make decisions. It often gets treated as a by-product, but I think it’s the most important part, and it will matter even more as AI becomes part of how we design and run games.

Sponsors usually come to a wargame wanting an answer. Which option is best? What will the other side do? Sometimes you get useful signals on those questions. What you reliably get is an honest look at the people in the room: what they assume, what they prioritise, how they trade things off, and where a team works well together or starts to fall apart. You don’t see that in a strategy document. You only see it when people have to make choices and live with the results.

Earlier this year I ran a strategy wargame for the leadership team of a global consumer goods company. Six teams played eight rounds, each with its own mandate and starting position. The market mechanics worked well, but what stayed with everyone was what the game showed about how they make decisions.

A few things stood out:

1. 17×team with least

   More resources made things worse. The team with the most was the only one to shrink, while the team with almost nothing grew seventeen-fold. As one player put it, with only three blocks they made sure every one had a purpose.

2. 7word strategy

   Simple strategies held. The team with a seven-word strategy rarely drifted. The team without one changed course constantly, and it showed in the results.

3. People defended what was theirs. Two teams kept putting resources into a market that wasn’t growing, round after round, because that’s where they started.

4. Losing changed behaviour. One team stopped reacting to its losses. The winning team kept repeating a simple line: let’s not panic.

None of that was really about confectionery (well… a little). It was about how that leadership team makes decisions, and in the debrief they recognised themselves. That’s what changes behaviour back at work. I’ve seen the same thing in Defence. The games were about force structure and capability on paper, but the most useful findings were the assumptions people didn’t know they were making and the places where it wasn’t clear who had authority to decide.

So where does AI fit?

Human games are expensive. They take weeks of design, days of senior people’s time and months of analysis, so you only get a handful of runs. That’s enough to see how people behave, but not enough to know whether an outcome was a real signal or just luck on the day.

This is where AI is extremely valuable. We can use agents and models to run a scenario thousands of times as a large-scale Monte Carlo analysis, varying the starting conditions, the opponent’s choices and the random events. That shows you the range of possible outcomes, which assumptions really drive the result, and where the biggest risks are. Before a game, it helps us put human players in front of the decisions that matter most. After a game, it lets us test what we saw. Was “less is more” a feature of the system, or just those six teams on that day? Run it ten thousand times and you’ll have a much better idea.

AI is also very useful for the grunt work: building an outside world that reacts to the players, drafting updates, and processing transcripts and decision logs in days instead of months. But the model can only tell you what’s possible inside the model. People tell you what people actually do. You need both.

The biggest challenge I see isn’t the AI itself. It’s whether people believe it. Sebastian Bae made a good point on a recent Wargaming Weekly episode with Rwizi Rweizooba Ainomugisha about AI adversary emulation struggling because you can’t trace why it made a decision. I’d add that credibility can go wrong in two ways. Some people trust polished outputs too much and treat a distribution as a prediction, when it’s only as good as the assumptions underneath. Others dismiss anything a machine produces, and good analysis goes unused. Both are human decision-making problems, which is a bit ironic given what we’re trying to expose.

Building that credibility takes work. That means:

If you can’t explain an AI move in the hotwash, it shouldn’t change your findings.

Some of the hardest decisions ahead will be about AI itself, made under real uncertainty with no history to draw on. Wargames are one of the few ways to rehearse those decisions while the stakes are still low. That’s why the human side of the game matters so much.

If you’re working anywhere near strategy, crisis response or AI governance, I’d be keen to hear where a game has shown you something about decision-makers that no other analysis could.

First published on LinkedIn, 20 October 2026.