Every enterprise AI conversation right now happens in the same phase, whether anyone uses the same term for it or not. Pilots get funded. New ones get greenlit faster than old ones get properly reviewed. The size of the AI budget gets read as a signal of intent rather than a cost that has to be justified line by line. Everybody I talk to who works inside one of these programmes describes some version of the same environment: permissive, well-funded, short on scrutiny.
The assumption people reach for, when I ask how long that lasts, is that the phase is temporary. A correction is coming. Call it the post-pilot era: the point where AI spending finally gets rationalized the way cloud spend eventually got audited, and RPA licences eventually got questioned once the first wave came up for renewal. I hear this from people who should know what they're talking about, CDOs and heads of transformation who have lived through at least one of those correction cycles already and assume this one will follow the same arc.
I'm not convinced the correction arrives on anyone's schedule, or that it arrives at all. Cloud spend and RPA licences both had something the AI pilot spend doesn't: an external party who eventually forced the question. A CFO reviews an invoice that's grown 40% year on year and asks what it's buying. A manager flags a renewal nobody can justify. An AI pilot that hasn't reached production yet doesn't generate that kind of pressure, because it doesn't show up as a recurring line item anyone has to defend. It shows up as a strategic initiative, which is a different budget category with a much longer leash.
There's also nobody with an obvious reason to be the one who ends it. Cancelling an AI initiative someone else started, in front of a board that funded it as a strategic priority, is a specific kind of career risk with no clean offsetting reward. A live pilot presents itself well in a steering committee deck regardless of what it costs or whether it works. A cancelled one appears like an admission of failure, and lessons learned don't get promoted up the organization. So the rational move, for almost anyone with the authority to make the call, is to let it keep running rather than be the name attached to shutting it down. The incentive against ending it doesn't have to stay implicit. At one company I heard about recently, it's written directly into somebody's job.
A recent conversation about a European industrial group made the mechanism concrete rather than theoretical. The Chief AI Officer's metric is tied to how many pilots he launches in a given month. Not how many reach production. Not how many get shut down because they didn't work. Pilots launched. Whatever he privately believes about which of them deserve to survive, the number his own performance review runs against only moves in one direction when a new one starts. Cancelling one, however overdue, costs him against his own target and returns nothing. Whether or not anyone intended it, the company built a KPI that keeps the open bar running: a personal metric rewarded for starting pilots, with nothing on the other side of the ledger for ending them.
None of this is unique to that one executive or that one company. It's a sharper version of something I wrote about in an earlier post: the org chart rewards the person who ships a new system and has no equivalent reward for the person who deletes an old one. What the pilot-count metric does is turn that general absence of reward into an explicit incentive running the opposite direction, with a number attached and a name for the metric.
A KPI like that also guarantees a harder question never gets asked, because it isn't built to notice the question exists. Nobody measured on pilots launched has a reason to ask whether a given process needed a model in the first place, or whether a deterministic system would have solved it for less money and fewer surprises. Finding that answer would only cost them, so it doesn't get looked for.
So I'd stop assuming the post-pilot era arrives on its own, the way the last two correction cycles did. Nothing about this one guarantees an external party who eventually forces the question. If it happens at a given company, it will be because somebody chose to be measured by the pilots that got shut down instead of the ones that got launched: a decision, available to exactly one person at a time, who currently has no reason to make it. Worth checking, before assuming otherwise, how the people running your own AI programme are actually being scored.
That's the eighth pattern.