AI should be a mech suit, not a replacement

Think about a mech suit from any film you like. A person climbs in, and suddenly they can lift a car, run at forty miles an hour and punch through a wall. But every move the suit makes starts with a move the pilot makes. Take the pilot out and it’s just a very expensive pile of metal standing in a hangar.

That’s what I think AI should be for people at work. Not a robot that does the job while you watch. A suit that makes you stronger at a job you’re still doing.

Two kinds of AI tool

Once you see this, you see it in every AI product.

The autopilot takes the job off you. It writes the report, summarises the meeting, decides the priority, drafts the strategy. You just hit approve. It feels amazing for about three months. Then you notice you can’t really do the job anymore, and you’re not sure the AI is doing it well either, because you’ve stopped being able to tell.

The mech suit takes the heavy lifting and leaves you the steering. It gathers what you need, asks the questions you’d forget, remembers what happened last time and makes the boring bits disappear. But the judgement is still yours. You come out of it better at your job, not worse.

In the first two pieces in this series, I went through the evidence that autopilot has a cost: people stop thinking critically about what they trust, skills fade, and everyone’s ideas start to look the same. The mech suit is the answer to all three.

The amateurs who beat the grandmasters

My favourite story about this is from chess. After Garry Kasparov lost to IBM’s Deep Blue in 1997, he got interested in what happens when people and computers play together, a format that came to be called advanced or freestyle chess.

In 2005, an online freestyle tournament let teams use any mix of people and machines. Grandmasters entered with powerful computers. The winners were two amateurs from New Hampshire, Steven Cramton and Zackary Stephen, with modest ratings and three ordinary PCs. They didn’t win because they had the best machines or the best chess brains. They won because they’d worked out the best way of working with their software: when to trust it, when to check it against another engine, when to go with their own read.

Kasparov’s conclusion has stuck with me ever since. A weak human plus a machine plus a better process beats a strong human plus a machine with a worse process. The process is the thing. That’s the mech suit.

This isn’t a new idea

The people who invented personal computing saw it this way too. In 1962 Douglas Engelbart wrote a report at the Stanford Research Institute called “Augmenting Human Intellect”. The whole point of computers, for him, was to make people better at solving hard problems, not to solve them for us. Steve Jobs later called the computer a bicycle for the mind. You still pedal. You just go a lot further.

Somewhere along the way, a lot of AI products forgot that. The pitch became “AI that does it for you”. I understand why. It demos brilliantly. But a bicycle that pedals itself is a moped, and nobody ever got fitter riding a moped.

What it means for a company

Inside a company, the autopilot shows up as AI that writes everyone’s updates, scores everyone’s goals and tells the leadership team what to prioritise. It saves a lot of time. It also hollows out the one thing a leadership team is for: thinking hard together about what matters and deciding.

The mech suit version looks different. The AI does the chasing, the collating, the reminding and the remembering, all the coordination admin that eats a founder’s week. Then it hands the team the questions that matter, with the facts laid out, and gets out of the way. People decide. I’ve made that case in full in The machine asks, people decide.

Pick your side

I’ll be honest about where I stand. I think autopilot AI is going to make a lot of companies worse while making them feel faster. Their people will lose the muscle, their ideas will drift to the average and they’ll find out the hard way when the market shifts and nobody inside knows how to think it through.

The companies that win will be the ones that put their people in the suit. Same AI, very different design. That’s what we’re building with Clearahead, and the last piece in this series is about exactly how: Why AJ asks instead of answers.

Questions people ask

What is the difference between augmenting and automating people with AI?

Automating means the AI does the work instead of the person, and the person’s skill is no longer needed. Augmenting means the AI makes the person stronger at work they still do themselves: it handles the heavy lifting, but the person still steers, judges and decides.

What is a centaur in AI?

A centaur is a team of a person and an AI working together, a term from advanced chess. In a 2005 freestyle tournament, two amateur players using ordinary computers beat teams of grandmasters with machines, because their process for working with the software was better.

How can you tell whether an AI tool is making your team better?

Ask four questions. Would your team be better at the job if the tool was switched off tomorrow? Does it ask you before it tells you? Can you explain why its output is right? Does the result still sound like you? If the answers are mostly no, it’s replacing your team’s thinking rather than amplifying it.