Buying Copilot licenses doesn't make your team AI-native any more than buying gym memberships makes people fit.
I've said this to a few engineering leaders recently and gotten the same reaction: a brief uncomfortable pause, then "yeah, that's exactly what's happening." They bought the gym membership. Nobody's going to the gym.
This is the central problem with how most organizations are approaching AI adoption, and I think it's worth being direct about it: the bottleneck isn't access to tools. It's behavior.
Why the tool-purchase instinct is so strong
When a technology wave hits, buying things is the path of least resistance. It's a clear action with a measurable outcome. It goes in the budget. It satisfies the board question about whether you're "investing in AI." It generates an announcement.
Behavior change is slower, messier, and harder to put on a slide. You can't buy it in one procurement cycle. It requires sustained attention, repetition, and often process redesign that nobody volunteered for.
So organizations do what's easy and call it done. They've "invested in AI." The reality on the ground looks nothing like the announcement.
What behavior change actually requires
I've watched this play out in teams across different functions — engineering, product, support, sales. The pattern is consistent.
When someone first gets access to an AI tool, there's a novelty phase. They try it for a few things. Some work well, some don't. Then the novelty fades, and unless something actively sustains the habit, usage drops back toward zero.
For the habit to stick, three things need to be true:
There has to be a clear win early. If the first few uses don't save meaningful time or produce noticeably better output, people don't come back. The onboarding experience matters more than most companies think. What's the one workflow where this tool makes someone's Tuesday 30% faster? Find it. Start there. Build out from the early win.
The workflow has to change, not just include a new option. This is the redesign problem. If using AI is optional and bolted onto an existing process — something you can do rather than something that's built into the flow — most people won't do it consistently. Adoption accelerates when the process is redesigned so that the AI-assisted version is the default path, not the alternative one.
There has to be social proof and normalization. Behavior is socially regulated. If senior engineers are vocally skeptical and the team's culture treats AI usage as a shortcut or a cheat, adoption stalls regardless of tool quality. If the lead engineers are openly sharing what they're using AI for, talking about when it helped and when it didn't, and treating it as a normal part of the workflow — that matters enormously.
The middle state most teams are stuck in
There's a frustrating middle state where teams have tools, some people use them, but it's uneven and fragile. A few power users get real value. Most people use AI occasionally for low-stakes tasks. Nobody has really changed how they work.
This feels like progress but it's actually an unstable equilibrium. The teams that make real gains go further: they build the AI tools into their processes deliberately, they create internal sharing around what works, they hold themselves accountable for actually changing workflows — not just providing access.
The teams that stay stuck treat adoption as an infrastructure problem. "We have the tools. It's on people to use them." That's the gym membership fallacy.
The manager's role is different than you think
Engineering managers often think their job in AI adoption is to evaluate and procure the right tools and then get out of the way. That's maybe 20% of the job.
The bigger 80% is:
- Modeling usage themselves. If you're not using AI in your own work and talking about it, you're signaling (even if you don't mean to) that it's optional.
- Creating shared context about what's working and what isn't. Run a retro specifically on AI workflows every few sprints. Make it normal to talk about this.
- Redesigning processes. Not just adding AI as an option, but asking: if AI is available, what does the right version of this workflow look like now? Then building that.
- Protecting the time to learn. Behavior change requires experimentation. If your team is under enough sprint pressure that there's no slack to try new approaches, they won't. Full stop.
The honest question to ask your team
Don't ask "are people using AI?" Ask: "Would it feel weird and inefficient to do this task without AI now?"
If the answer is yes for a meaningful set of workflows, you're making progress. If the answer is "it depends on the person" or "probably not," you have work to do — and that work is about habit and process, not tools.
This is the uncomfortable truth about AI adoption: the tools are mostly table stakes at this point. Everyone has access. The differentiator is whether your organization has actually changed how it works — and most haven't.