The conversation goes the same way every time. We propose a sprint. The leader nods politely. Then, almost as an aside: "We tried something like this in 2023. It didn't really land."
That sentence carries a lot of weight in the room, and it should. The pattern behind it is real. A vendor pitched a capability, a cross-functional team was assembled to evaluate it, a six-month pilot was scoped, a workshop or two happened, a proof-of-concept materialized in a sandbox, and then nothing made it into production. Eighteen months later, the pilot is a slide in a deck and a small line in next year's budget marked "innovation."
If that's the experience, the lesson the room walked away with is: GenAI isn't ready.
It's the wrong lesson. The premise wasn't wrong. The engagement shape was.
Why first-wave pilots stalled
The pilots of 2023 weren't sabotaged by the technology. They were sabotaged by the way the work was packaged.
The unit of work was too big. "Let's explore how AI could transform our X" is not a sprint. It's a discovery process pretending to be a delivery one. The team spends three months learning what the tool can do, two months arguing about which use case to pursue, and one month building a demo that proves the obvious. By the time anything ships, the original sponsor has moved on and the demo lives in a sandbox nobody can extend.
The team shape was wrong. First-wave pilots were staffed for debate, not delivery. A product manager. Two enterprise architects. A vendor solutions engineer. A change-management consultant. A workshop facilitator. By design, that team is good at producing recommendations. It is not built to ship working software. The team that diagnoses is rarely the team that executes, and the handover between them is where most pilots quietly died.
The vendor was selling a capability, not a solution. The pitch in 2023 was "look what our model can do." The buyer needed "look what your business will do differently next month." Those are different products, and the second one is far harder to deliver than a capability demo. Most vendors didn't even try.
The handover never happened. The pilot lived in a vendor environment, on a vendor stack, with vendor credentials. When the pilot ended, the work ended with it. There was no production code, no runbook, no internal owner. The thing that worked in the demo could not be operated by the company that paid for it.
2023 didn't prove the model was weak. It proved that capability demos run by debate-shaped teams under vendor control don't make it into production.
What changed
Two things, both quiet.
The unit of work got smaller and sharper. A working sprint today is one identified problem, one named owner on the client side, one fixed scope, one outcome, one deadline. No exploration phase. No use-case workshop. The diagnosis happens before the engagement starts, often in a single conversation. By the time the sprint opens, both sides already know what is being built, who will use it, and how success will be measured. That clarity at the start is what most 2023 pilots never had.
The same people diagnose and ship. The handover failure mode is gone because there is no handover. The team that maps the process is the team that writes the code, configures the pipeline, hands the system to the operator, and stays available for the questions that come up in week three. The shape of the engagement makes drift impossible.
None of this is about the model getting smarter, though it has. It's about the engagement architecture finally catching up to what the technology made possible two years ago.
How to test it again
If you are the leader who carries the 2023 pilot in the back of your head, the right move is not to wait another year. It is to re-run the test with a sharper scope and a different team shape.
Pick one operational problem that has a clear before and after. Not "AI strategy." Not "data transformation." A specific bottleneck a specific team would notice if it went away. A weekly report that takes too long. A backlog nobody can prioritise. A form your customers abandon. Anything where you can describe success in a sentence.
Then scope the engagement so that the outcome lands in production in weeks, not quarters. One sprint. Fixed price. Working output, fully documented, handed to your team at the end. If the sprint delivers, you have a result you can extend. If it doesn't, you have spent a small fixed amount and you know more about what your business actually needs than any number of workshops would have told you.
The 2023 pilot deserves the credit it has earned with you. It made you a sharper buyer. The next test should be designed for the buyer you became, not the one you were when the first pilot started.