Everyone wants the 10x. Very few teams get it, and the ones that do rarely got there by buying a better tool.

Look at the research and the spread is almost comical. A 2025 study found AI adoption cut PR review cycles by roughly a third for some teams. A METR randomized trial found experienced developers without proper scaffolding actually slowed down by 19%. Same tools. Opposite outcomes. The variable isn't the model — it's everything around it.

Here's the uncomfortable truth I keep landing on: AI amplifies what already exists. Point it at a team with strong foundations and senior judgment and you get leverage. Point it at chaos and you get faster chaos.

Three habits the winning teams share

1. They define "done" explicitly. An AI agent will happily mark work complete the moment the code compiles. High-performing teams don't let it. They bake real completion criteria into the workflow — tests pass, integration holds, security scans clear, it deploys to staging — as formal gates the agent has to clear. "It runs" is not "it's done."

2. They treat agents like team members, not utilities. I watched a CTO change an agent's label from "product manager" to "scrum master" and its behavior shifted — because it had absorbed the internet's definition of those roles. Roles carry expectations. Design them on purpose.

3. They keep experienced humans in the loop. A 2026 ICSE study found developers using AI produced substantially more code and deleted significantly more of it. Someone has to catch the confident-but-wrong decisions before they ship. One agent I know of couldn't reach a database, so it decided — on its own, unprompted — to spin up a brand new PostgreSQL instance. A senior engineer caught it before production. That's the judgment you're paying for.

The tiers

With proper foundations — CI/CD, testing, documentation, senior review — most teams can reach a real 3x to 5x. Layer on full agent orchestration with defined roles and human gatekeeping, and 5x to 10x is on the table. But the ceiling moves with your discipline, not your license count.

And the downside is real. A major cloud provider's internal pressure to adopt AI contributed to a multi-hour production outage in late 2025 when AI-generated code bypassed proper review. Speed without discipline isn't productivity. It's liability with a shorter fuse.

For leaders

Don't mandate adoption — establish the foundations first. Use AI to build scaffolding faster, under expert direction. And above all, protect your experienced engineers, because they are the multiplier. AI excels at boilerplate; integration-heavy mature systems demand more care, not less.

The fastest teams aren't the ones winning. The teams winning are the ones with strong foundations paired with judgment — and AI turned all the way up on top of both.