I want to start with a chess story, because it explains hybrid intelligence better than any slide I've made.
The tournament nobody expected
In 2005, an online chess site ran a "freestyle" tournament. Anything was allowed. You could enter as a grandmaster, as a computer, or as any mix of people and machines working together.
Everyone expected one of two winners. Either a grandmaster using a strong computer, or simply the strongest computer.
Neither won. The winners were two amateur players from New Hampshire, running a few ordinary laptops.
Garry Kasparov, who'd lost to Deep Blue eight years earlier, wrote about it later. His conclusion is one line, and I think every leader deploying AI should read it twice:
"Weak human + machine + better process was superior to a strong computer alone and, more remarkably, superior to a strong human + machine + inferior process."
Not the best player. Not the best machine. The best process between them.
Hybrid intelligence is a process, not a headcount
Most companies I meet are still asking two questions. Which model is the best? And how many people do we still need?
Both are the wrong question. The 2005 result says the winning variable is neither the AI nor the people. It's how they work together.
That's what hybrid intelligence means to me. It isn't a ratio of humans to agents, and it isn't "AI, with a person watching". It's a designed way of working where:
- Each side knows what the other is good at. The amateurs didn't out-think the grandmasters. Kasparov credited their skill at steering and "coaching" their machines: knowing where to point them, and when to trust them.
- The work is split on purpose. Not "AI does it, a human approves it", but each part of the job goes to whoever, person or agent, does it best.
- They disagree in the open. When the machine and the person see it differently, that's not a failure. That's where the best decisions come from, as long as someone owns the call.
- The process gets better every time. Every game, every case, every decision teaches the team something, and that learning stays with the team.
A brilliant model inside a bad process loses. An average model inside a good one wins.
"But chess engines don't need people any more"
That's true, and it's worth being honest about. In chess, engines got so strong that a person now adds very little.
But chess is a closed board. Every piece is visible and every rule is written down.
A company is the opposite. Half of what matters isn't written anywhere: the customer's mood, the history with a supplier, the reason a rule exists, what a new regulator actually cares about. The board is never fully visible, and the rules change every quarter.
On that kind of board, the centaur doesn't retire. It's the only player that can see the whole game.
What this means on Monday
If you're rolling out agents, spend less time comparing models and more time designing the process around them:
- Write down who does what, agent or person, for one real workflow.
- Decide where they hand work to each other, in both directions, and why.
- Decide who owns the call when they disagree.
- Make sure every decision and its reason is kept, so the next one is better.
That's the work we do at LeafCraft: one platform where AI agents and people work as one team, with clear jobs, handoffs both ways and every decision on the record.
And it's what this newsletter will dig into, one real problem at a time. Subscribe if you're working through the same questions. I'll share what we're learning, including what doesn't work.
Be honest: in your company today, is AI mostly a better tool for individuals, or is it part of a designed team process? Tell me in the comments. I read and answer every one.
Originally published as edition 1 of my LinkedIn newsletter, The Hybrid Enterprise. I'm the founder of LeafCraft, the Hybrid Intelligent Enterprise platform, and we build the LeafMesh ADK (pip install leafmesh).
Source: Garry Kasparov, "The Chess Master and the Computer", New York Review of Books, Feb 2010.
Top comments (0)