Six weeks ago I was working with a team that bought all the AI licenses at once. Sound familiar? Everybody was in the room for the demo, and it looked absolutely incredible. Code suggestions started flying. Tickets started closing faster. A real buzz was building.
Fast forward to today. They're shipping the same confusion they've always shipped, just faster. The bugs come quicker now. The rework comes quicker now. Everything got faster.
But nothing actually got better.
If that scene sounds familiar, you're not behind, and your team is not broken. You just ran into something a lot of leaders are discovering this year: AI didn't fix the system. It turned up the volume on whatever the system already was.
AI Turned Up the Volume
Here's what usually happens. A leader sees a great demo, signs the contract, and tells everyone to start using the AI tool. For a couple of weeks it feels like magic. I've been there too. Then the cracks show up.
Picture a team that never wrote clear acceptance criteria. Their user stories were always a little fuzzy. People figured it out in hallway conversations and Slack threads. It was slow, but it sort of worked. Now hand that team an AI assistant that writes code very fast. It writes code against fuzzy requirements just as fast. Did that help anybody?
No. Now they're producing the wrong thing three times quicker than before. The confusion didn't go away. It just scaled.
AI Is an Amplifier, Not a Strategy
What most vendors won't tell you is that AI is not a strategy. It's an amplifier. If your decisions are clear, AI makes those decisions faster. If your decisions are muddy, AI makes muddy decisions louder.
Think about a microphone. A microphone doesn't make you a better singer. It makes you a louder version of what you already are. If you're off key, now everyone in the building knows it. That's the whole game. The tool is honest, painfully honest, and it shows you what your operating system actually does, not what your strategy deck says to the board.
What the Data Actually Says
This isn't just my opinion. If you've followed me for a while, you know I love data about how teams actually perform. Every year a research group called DORA studies exactly that. DORA's 2025 report focused specifically on AI-assisted software development. They surveyed around 5,000 technology professionals and did hundreds of hours of interviews.
The headline finding is the one I want you to sit with: AI is an amplifier. It magnifies the strengths of high-performing organizations and it magnifies the dysfunctions of the struggling ones.
90% Adoption, 30% Low Trust
Two numbers make it real. Adoption is almost universal now. Roughly 90% of the teams they studied are using AI in some form. But trust is shaky. About 30% of practitioners said they have little or no trust in what they're getting back.
So almost everybody is using it, but a third of them don't even trust it. That gap is where the trouble lives, and it matches what I see in the field.
Speed and Fragility Rise Together
Here's the part that catches leaders off guard. DORA found that AI tends to push throughput up and instability up at the same time. You ship more, but you also break more. Speed and fragility rising together. That is not productive.
One honest caveat: this is correlation, not a controlled experiment. DORA is showing patterns across thousands of teams, not proving cause and effect. So hold the causal language loosely. But the pattern is strong, and it consistently matches what I see in the field.
The System Is What Separates the Winners
So what separates the teams that win? The surrounding system, not the tool. DORA identified a set of organizational capabilities that decide whether AI helps or hurts. And where are organizational capabilities born? With leaders.
But leaders don't have all the answers. Remember the lesson from Lean and Toyota: the employee with the wrench usually knows best how to use the wrench. So push the innovation down to the front line. Involve your teams in setting the policies.
Here's the short list of what a healthy AI ecosystem actually needs:
- A clear AI policy
- A healthy data ecosystem (garbage in, garbage out)
- Internal data that AI can actually reach
- Good version control
- Working in small batches
- Staying user-centered
- A quality internal platform
Read that list again in your head, and notice something. Not one of those is an AI feature. Every one of them is part of your system. The tool is the same for everybody. The system is what differs.
BCG's 10-20-70 Rule
There's a second source worth pairing with this. Boston Consulting Group describes AI transformation as a workforce transformation, not a technology purchase. They use a rule of thumb they call 10-20-70. Roughly 10% of the value comes from the algorithms, about 20% from the data and technology, and around 70% from the people and the process change.
BCG sells consulting, so take the exact split as advisory, not gospel. But the shape of it is hard to argue with. The smallest slice is the tool. The biggest slice is how your people work and decide.
Put DORA and BCG side by side and you get the same message from two directions. The tool is the cheap part. The system around it is the expensive part, and it's the part that actually determines your result.
If you bought the tool and skipped the system work, you didn't buy a transformation. You bought a louder version of the problems you already had.
Read Your Results Like an X-Ray
So what do you do with that? Stop reading your results as a scorecard and start reading them as a diagnostic. When you roll out AI and something gets dramatically better or dramatically worse, that's not just a result. That's information about your system. The tool is acting like an X-ray, showing you the bones underneath.
Let me make it concrete with two teams getting the exact same tool.
Team A has clear ownership. They write small, well-defined stories. They have solid tests and a clean deployment pipeline. Give them an AI assistant and it drafts the code, good work flows downstream, throughput goes up, and quality holds. The amplifier made their strengths louder.
Team B is what we usually see. The team looks busy but works in a tangle: big batches, weak tests, unclear ownership, and a deployment process held together with hope. Give them the same assistant and they generate more code faster on top of a shaky foundation. Defects climb. Rework climbs. The amplifier made their dysfunction louder.
Same tool, opposite outcomes. The difference was never the AI. It was the system the AI landed on.
The One Question That Does the Work
For our premium email subscribers this week, I built something I'm calling the AI Amplifier Diagnostic: a short set of questions that reveal what your specific rollout is actually amplifying. I won't read the whole thing here, but I'll give you the sharpest one, because it does most of the work.
What has gotten louder since you adopted AI?
Sit with that a moment. Don't answer with the official version. Answer with the real one. Maybe meetings got louder because now there's more half-baked code to argue about. Maybe a turf war got louder because the tool exposed who actually owns a decision. Maybe a quality problem got louder because you can finally see how fast you're producing defects.
Whatever surfaced, that's your X-ray. The tool didn't create it. The tool revealed it.
And here's the leadership step that follows. You don't blame the tool. You don't blame the team. You go fix the thing that got loud. If acceptance criteria are the problem, tighten them. If the platform can't carry the gains past one developer, invest in the platform. If trust is low because nobody reviews the AI output, build the review loop.
The real issue is that most teams won't. They'll blame the tool, churn the vendor, and buy a different one next quarter. The brave teams read the result and name what's broken. That second group is who pulls ahead this year. As leaders, that usually means we make it a priority and give the team permission to tackle the process change, even though it takes time away from delivery.
A 15-Minute Exercise for Your Team
Here's the step for the week. It takes about 15 minutes. Pick the one task or workflow that AI has sped up the most this month. Just one, the thing where everybody agrees it got faster. Get the team in a room or on a call, ask two questions, and write the answers where everybody can see them.
First: Where did the saved time go? Did it turn into better work, more learning, more breathing room? Or did it just turn into more output nobody asked for?
Second: What got noisier since we started using this? More bugs, more rework, more arguments, more confusion about who decides what?
That's it. You're not solving anything yet. You're just naming what the tool has amplified, out loud and together. Naming it is the part most teams skip. Once it's named, you can decide what to fix. You'll have done more honest diagnostic work in 15 minutes than most organizations do in a quarter.
AI Is Your X-Ray, Not Your Treatment Plan
I'll leave you with one line. AI is your X-ray. It is not your treatment plan. It shows you what's there. Fixing it is still your job.
Most people are stuck in the trap of treating AI as a tool. The move that separates the teams pulling ahead is building the system and the organization around it, the checks and balances that let AI make you better instead of just louder.
If you want help building that operating system, that's the work we do at Big Agile. Explore our classes and coaching at big-agile.com and let's turn your AI rollout into a real transformation instead of a louder version of the old problems.