Organizations don’t fail at transformation because they picked the wrong tool. They fail because they deploy every tool the same way — mandate first, accountability never. They did it with Agile. They’re doing it with AI. The pattern is identical and the cost is the same. Only the timeline is faster.
That’s not a process failure. It’s a control reflex. When leadership doesn’t trust uncertainty, they force structure onto work they don’t yet understand. The estimate creates the appearance of predictability. The uncertainty doesn’t go anywhere — it goes underground, where it compounds until it surfaces as a missed deadline nobody can fully explain.
The AI version of that reflex is running right now. In Agile it was the forced estimate — sized before the spike, committed before the work was understood, wrong before the sprint started.
The mandate arrives from the top. AI in the SDLC. AI in quality testing. AI in roadmap planning, backlog refinement, documentation. All of it, now. The tools get deployed. The workflows get modified. Teams start using them because using them is no longer optional. Some people use them well. Some paste whatever comes back. Some quietly gave up after the first bad output. Leadership sees adoption metrics. They don’t see the difference.
Nobody has answered the accountability question.
The rollout doesn’t arrive as a directive. It arrives as a calendar invite. Here are the demo sessions. Here is the tool. Here is the dashboard showing who is using it and how often. The question the dashboard can’t answer — and nobody is asking — is what the output is supposed to look like when someone uses it correctly. The metric exists before the standard does.
An engineer submits a prompt. The AI writes a test. The test passes review. It ships. Three months later, something slips through that the test was supposed to catch. It was too generic — surface coverage that looked complete and wasn’t. Who owns that? The engineer who wrote the prompt? The AI that wrote the test? The team that shipped it? Nobody has a clear answer, because nobody asked the question before the rollout.
That gap isn’t unique to testing. It shows up wherever AI output enters a delivery workflow without a defined owner.
I’ve seen what a test suite nobody trusts actually costs a team. Engineers run it because the process requires it. They don’t rely on it. A green build stops meaning anything. The suite grows larger, the signal degrades, and the organization continues to believe its quality coverage is intact because the numbers look right. The checks are happening. The catches aren’t.
Now add AI writing those tests at volume. The suite grows faster. The coverage looks better. The signal degrades faster — and with fewer people close enough to notice, because the whole point of the automation was to reduce the human touchpoints.
The nuances don’t wait. They accumulate. And when they surface, the AI gets the blame — because the system that deployed it was never designed to examine itself.
The speed is new. The reflex is not.
Want the Experiment-Driven Agile Retrospective Toolkit?
If you’d like the Toolkit, reach out and I’ll send details (what’s included, pricing, and how teams use it). Or subscribe for new posts and updates.