The AI CEO experiment is testing the wrong thing

A Toronto startup called Skyfall AI is going to spend up to a million dollars buying a small software company, install an AI as its CEO, and see what happens.

It’s promised to double the revenue inside six months and to publish the failures alongside the wins.

I want to be clear that I think this is a good idea. Not the AI CEO part — the publishing-the-failures part.

Enterprise AI has spent three years making unfalsifiable claims. A company volunteering to be proved wrong in public is doing something the rest of the sector has carefully avoided.

I should also say that I’m not the resistance here. I automate everything I can. AI now handles much of the administration I resent, so I spend less time on invoices and calendar Tetris. That’s exactly what the technology is for: the work machines do better than we do, the work we find tedious, and the work that is frankly beneath us.

Which is precisely why I think Skyfall has chosen the wrong thing to automate.Which is precisely why I think Skyfall has chosen the wrong thing to automate.

Not because AI can't make decisions. Quite the opposite; it can. The mistake is assuming that making decisions is where the value of a CEO lies.

The CEO job was never operations

Sam Pasupalak, Skyfall's founder, reckons more than half his time goes on operations: meetings, budgeting, coordinating teams, running campaigns.

Hand that half to the machine, the reasoning goes, and he gets the other half back for strategy. Every CEO I’ve worked with believes this: that underneath all the scheduling there’s a real job, and the meetings are barnacles on the hull of it.

But look at what’s actually on that list: Coordinating teams isn’t operations. Sitting in the budget meeting where four departments find out whose plan survived isn’t operations either. It is the job.

The CEO's role has never really been about running things; it’s about mobilizing people behind a common goal, making the decisions nobody below you can make, and being seen to make them.

Those things simply happen to arrive in your calendar looking like admin, and the resemblance is the trap. Pasupalak looked at a week full of leadership and misfiled it under logistics on the grounds that it turned up as meeting invites.

Corporate life has accumulated an extraordinary collection of euphemisms for communication: "stakeholder management", "alignment", "buy-in", "change management", "socializing decisions".

Strip away the vocabulary and they all describe the same thing: the process by which one person's decision becomes several hundred people's shared understanding.

Skyfall’s decision to buy a business worth under $1m is tractable precisely because it contains little organization. There’s no middle layer mistranslating decisions on their way down. No team that stopped following a policy because nobody explained why it existed. Nobody concluding, in a meeting the CEO doesn't attend, that this new initiative will probably blow over like the last one.

The hard part was never the decisions. It’s that a decision, once made, has to survive contact with people who didn't make it—and might not like it.The hard part was never the decisions. It’s that a decision, once made, has to survive contact with people who didn't make it—and might not like it.

Proving AI can run a company that has no internal communications problem is testing your bridge design on a stream.

Skyfall's explanation for why it will succeed is a thing it calls an Enterprise World Model: a system that simulates how decisions ripple through an organization.

But revenue, churn, tickets and conversion are the exhaust; they tell you what came out of the organization, not what happened inside it. They don't capture what people believe, what they understood, or which processes everyone ignores. Organisations run on shared understanding that is mostly unwritten and almost entirely unmeasured.

I watch a smaller version of this every week. People use AI well, privately, and the organization never finds out. Sometimes there’s no useful channel in which to share what they've discovered. Sometimes there’s no incentive to. If finishing your work in half the time simply means being given twice as much of it, silence is the rational response.

The result is a company full of augmented individuals and unaugmented teams, each reinventing the same prompts, solving the same problems and making the same mistakes. Individual capability has increased. Collective capability hasn't. That only grows when what one person learns becomes available to everyone else.

Skyfall's CTO Kaheer Suleman said: "Humans should always remain responsible in some capacity."

What capacity? Accountability isn't just governance; it's how organizations learn. Information has to travel upwards to somebody with both the authority and responsibility to change course. Separate those two things and you break the loop. The machine makes the decisions. The human carries the consequences.

What Skyfall’s experiment can’t test

For what it's worth, I think there's a decent chance the experiment works.

If an AI can run a small SaaS business and double its revenue, the finding won't be that AI can be a CEO.

It'll be that running a small SaaS business was mostly administration all along, and we'd dressed it up as leadership.

What it won't prove is that leadership scales the same way. Running a business stops being a decision-making problem long before it stops being a communication problem. There comes a point where the scarce resource is no longer the quality of the decisions being made, but the quality of the shared understanding around them.

At that point the CEO's job changes. It becomes less about deciding and more about ensuring that decisions survive contact with hundreds of people who weren’t in the room when they were made.

That’s the threshold the experiment isn’t built to test. Once a business reaches that point, intelligence is only one ingredient. Decisions have to travel downwards. Experience has to travel back up. Somewhere in the middle, people have to understand each other well enough to keep moving in the same direction.

That's why this is a communication story, rather than an AI one. The experiment may tell us something fascinating about decision-making. It won’t tell us much about leadership.

Because leadership isn't what happens when a chief executive reaches a conclusion.

It's what happens next.