Nobody Knows Which AI Tool Wins, So Optimise for Exit, Not Prediction

Every SME right now is having some version of the same meeting: which AI tool should we standardise on? The unspoken assumption is usually that whatever's already bundled into the office suite will win by default — Copilot, Gemini, take your pick. People trust what's already there. Nobody wants to have the discussion.

Except the discussion is already happening, whether companies plan it or not. Shadow AI use is real, and right now Claude is genuinely popular as a company tool — not because it's built into anything, but because people picked it. That undercuts the whole "default wins by inertia" theory before it's even been tested.

This is the centralise-vs-decentralise IT debate wearing a new outfit. Some companies will want one general tool everyone collaborates in. Others will end up with a mix — a company-wide tool plus specialised ones per team, the same way office suites coexist with a dozen team-specific SaaS tools today. Nobody knows yet which shape wins, or even which tool wins within that shape.

So the honest answer to "which AI tool should we pick" is: it's very hard to make predictions of what are good choices or even criteria right now. Nobody knows who's gonna win this race and who will disappear.

Which means the research-your-way-to-the-right-answer approach is solving the wrong problem. The market is too young for a correct answer to exist. What you can control is how expensive it is to be wrong.

The most important thing is making sure it's not too painful to switch later on. Don't build irreversible dependencies into a tool because it happened to be free and pre-installed. Optimize for exit, not for prediction.