The phase you're stuck in without knowing it

Something odd is happening in professional services right now. Adoption of AI is high and climbing. The financial return from it mostly isn't.

Clio's 2026 Legal Trends Report found that 71% of solo lawyers and 75% of small firms have adopted AI in some form. Most solo and small firms are using the technology already. Fewer than one in three can point to a number that moved because of it.

This isn't a technology problem. The tools these firms are using are, by any reasonable measure, good. The gap is somewhere else, and it's worth being precise about where.

Most businesses using AI today are somewhere in the first two of three stages, and don't realize a third one exists.

Stage one is access. Someone on the team has ChatGPT, or Claude, or Copilot open in a tab, and it's genuinely capable of helping with real work. This is where almost every business is today, and it's not nothing. It's just also not where the value shows up.

Stage two is a harness. The AI isn't confined to a chat window anymore. It's wired into a tool that lets it actually do things: search, write code, draft inside a CRM, chain a few steps together on its own. This is where the more adventurous businesses are, and it does feel like more is happening, because in a narrow sense it is.

Stage three is a role. This specific piece of the business is an AI employee's actual job, with defined boundaries, someone accountable for what it does, and a way to tell whether it's actually working. Almost nobody's there yet, and it's the stage where a recent AICPA and CIMA survey on finance and technology found the real friction: 88% of finance professionals said AI would be the most transformative technology trend they'd face, and only 8% said they felt very well prepared for it. That's not a capability gap. Firms that already believe the tools work are still telling their own trade body they aren't ready to use them properly.

Stages one and two both feel like progress, and they are, narrowly. But they're also where most of the noise sits without much of the return, because access and capability were never the actual bottleneck. The bottleneck was always whether anyone had said, precisely, what the job is. A model in a tab hasn't been given a job. Neither has a harness with no boundaries around what it may decide on its own. A role has.

If that gap sounds abstract, the next piece in this series makes it concrete: the specific, learnable signals that tell you a piece of your business is actually ready to move from "we have access to AI" to "this is a real role, and it's someone's job to be accountable for it."