The most reliable sign that a team is at Stage 1 is that they think they're at Stage 2.
I don't say that to be dismissive. It's a pattern I've watched repeat across organizations of different sizes, different industries, different levels of engineering sophistication. The visibility problem in AI adoption is real: it's genuinely difficult to calibrate where you are without an external reference point.
Over time, working with teams and studying the spread, I've mapped four stages. They correspond to the tiers the AI Adoption Score uses. Here's what each actually looks like — and more importantly, how teams move between them.
Stage 1: Bystander
A Bystander organization has the conversation but not the behavior. AI is discussed in leadership meetings. Someone has evaluated Copilot. The board has asked the question. But at the working level, almost nothing has changed.
The tells:
- AI tool usage is low and concentrated in a few individuals
- No formal process for AI, which in practice means no process
- The mental model is still "AI is a thing some people use for some tasks"
- Leadership talks about AI adoption as a future project, not a current one
The trap here is that Bystanders often feel like they're taking AI seriously because they're paying attention to it. Attention and adoption are different things.
What moves a Bystander forward: One clear, visible win. Not a pilot. Not a proof-of-concept. An actual workflow change that a real team is using, producing measurable output improvement, and talking about openly. The best Bystander transitions I've seen started with a single engineer or PM who redesigned one workflow with AI and shared the results. It gave permission and showed the way.
Stage 2: Dabbler
This is where most technology companies honestly sit right now. Dabblers have moved from conversation to tooling, and some people are genuinely using AI regularly. But usage is uneven, unmeasured, and unbaked into actual processes.
The tells:
- Some teams use AI extensively; others barely at all
- No shared language or best practices around AI workflows
- Workflows haven't really changed — AI is an option, not a default
- Leaders support adoption in principle but aren't modeling it themselves
Dabblers often feel like progress is happening because it is, in patches. The mistake is thinking the patches add up to organizational change. They don't, automatically. Without deliberate work to systematize and spread what the early adopters learned, you stay in this stage indefinitely.
Many organizations have been Dabblers for two years now. They will continue to be.
What moves a Dabbler forward: Systematization. Take what the power users learned and build it into the process for everyone. This means actual workflow documentation — not "AI is available" but "here's the new default workflow for this task, and here's why." It means creating venues for knowledge sharing (a Slack channel, a sprint retro item, a monthly internal demo). And it means managers starting to model usage themselves, visibly.
Stage 3: Operator
An Operator team has cracked the code on integration. AI is embedded in the actual workflows for most of the team, most of the time. It's not optional anymore — it's how the work gets done.
The tells:
- Engineers would describe "not using AI" for certain tasks the way they'd describe "not using version control" — technically possible, obviously not the right choice
- There are real process artifacts: standards, playbooks, shared prompt libraries, retro data on AI workflow performance
- Leaders use AI themselves and talk about it naturally
- The team has a clear-eyed view of where AI helps and where it doesn't — this nuance is important
The difference between Dabbler and Operator isn't about enthusiasm or tooling. It's about whether the process changed or just the option set changed. Operators redesigned workflows. Dabblers just made AI available.
What moves an Operator forward: Depth and governance. Moving to Frontrunner requires going beyond embedded usage to leading-edge practice — using AI in ways that genuinely differentiate output quality or speed, having governance mature enough to manage risk at scale, and pushing the frontier rather than following it.
Stage 4: Frontrunner
A Frontrunner organization is ahead of the curve, and they know it. AI isn't just embedded — it's strategic. The team is actively experimenting with new tools and techniques, sharing knowledge externally, and using AI to do things that would have been structurally impossible without it.
The tells:
- The team is producing content (blog posts, talks, internal frameworks) about AI workflows — they're thinking at a meta level
- AI is part of hiring, onboarding, and team structure decisions, not just tooling
- There's a governance layer that enables speed without recklessness: clear policies, audit trails, data handling standards
- The team is shaping how AI is used in the industry, not just keeping up with it
Frontrunner status is rare. I'd estimate fewer than 10% of engineering teams genuinely qualify. And the ones that do are rarely the ones with the loudest AI-adoption announcements.
Staying a Frontrunner: The main threat is complacency. The tools and techniques that made you a Frontrunner in 2025 are table stakes in 2026. Frontrunners stay Frontrunners by staying curious and by treating their own practices as evolving, not proven.
The honest diagnosis
Most teams reading this are at Stage 2 with aspirations toward Stage 3. That's fine — Stage 3 is achievable in 6-12 months with focused effort. But the effort is specific: workflow redesign, process documentation, leadership modeling, and knowledge systematization. It's not buying more tools.
If you want to know exactly where your team sits — and get a score across the five dimensions that matter — the AI Adoption Score will give you a concrete number and a diagnosis. It takes about 10 minutes. What comes back is a tier assignment and a set of actions specific to your current stage.
Most people are surprised by the results. Usually in the direction of "I thought we were further along."