Your engineers generate a lot of code quickly with AI, but it keeps missing requirements you laid out beforehand. Stories close a little faster than before, and you have a nagging feeling delivery should be much faster than it is.

If that sounds familiar, you probably aren’t operating AI-first. You got better at prompting.

The difference matters. AI-first is not a prompting skill; it’s a process-design decision. Your current process was designed for humans talking to humans — the handoffs, the meetings, and the documentation all exist to move understanding from one person to another. AI-first inverts that. Start from the question “what does the AI need to generate working software?”, design the process around the answer, and then build the human workflows to support the AI, not the other way around.

The leap is hard because it requires admitting that the process itself is the bottleneck — not the tools, and not the prompting technique. I’ve watched capable teams stall right there.

The foundation underneath the leap is agile in the original sense, not the ceremony that usually wears the name: short feedback loops, working software every cycle, collaboration over handoffs, responding to change over following a plan. Those principles were written for fast iteration. AI makes them mandatory, because an AI-speed pipeline feeding into quarterly planning rituals just piles up inventory faster.

For an objective standard, use DORA metrics. If AI isn’t moving lead time, deployment frequency, change failure rate, and time to restore, you’re still experimenting. The transformation hasn’t started yet.

There is a maturity progression here: writing code manually, copiloting, prompt engineering, context engineering, and then spec-driven development, where AI workflow tools help turn high-level problem statements into specs. Each step pushes more of the intent into the machine. No step lets you skip the foundation, though. AI bolted onto fake agile just makes the theater more expensive. On top of real agile, it changes what the team can do.

The teams making the leap aren’t the ones with the best prompts. They’re the ones who asked what the AI needs and rebuilt their process around the answer. For us, that question is where the whole redesign started.