What I've Learned Running an AI Program at a 6,000-Person Company

A few weekends ago I was reviewing the late-draft of a chapter for a book Vinay was writing. Vinay is my boss at OneDigital, where I run the AI program. He had asked me for a sentence for the section about how Tier One employees, the ones who use AI Coworkers every day without ever building one, should think about adoption. I gave him this:

AI adoption happens when you help people do today's work better and with more confidence, rather than when you ask them to believe in the future.

I gave him that sentence because it was the most honest summary I could write of what we have actually learned in eighteen months. It was also the part I would not have predicted in 2024 if you had asked me what enterprise AI deployment was going to look like. I had read all of the same vendor decks and analyst reports anyone in this work had read. The thesis those decks were selling was that AI deploys itself. Buy the licenses. Get out of the way. Watch productivity gains roll in.

What we have actually found is that AI does not deploy itself. AI does not take everything either. The middle, the boring middle, is where the work happens, and the boring middle is what almost nobody is writing about.

This is my attempt at writing about the boring middle.

Why both stories about AI are wrong

The two stories you hear most about AI in the enterprise are mirror images of each other, and both of them are wrong.

The first story comes from the vendor and analyst class. It says: AI is a horizontal productivity boost. Pick a model. Deploy it. Save 30% on cognitive labor. This is the story sold in keynotes. Ethan Mollick, in his Substack writing and in Co-Intelligence, has spent the last few years documenting what he calls the jaggedness of AI capability: it is superhuman on some tasks and useless on others, and the variance across people inside a workforce is wider than the variance across the model versions on offer. Whether your enterprise lands in the high-leverage or no-leverage cluster is not a function of which model you bought.

The second story comes from the doom and labor-displacement commentary. It says: AI takes everything. The agents replace the workers, the workers replace themselves, the firm reduces headcount until there is no firm. I have not seen that play out in any practice I work with at OneDigital. Not one knowledge-work function has been fully replaced. Plenty of routine sub-tasks inside knowledge work have moved to AI. The function around them has not.

The two stories share a quiet assumption: that the AI is the variable. Either the right AI saves the business, or the wrong AI sinks it. The truth is closer to what Andrej Karpathy was getting at with Software 3.0: when the AI is the foundation, the programming layer becomes the prompt and the workflow. The bottleneck moves from "do we have the right software" to "can we manage the new kind of teammate this software became."

That bottleneck is a people-and-process problem. It is the same bottleneck that determined whether enterprise CRM rollouts in 2014 succeeded or failed. It is the same bottleneck that determined whether ERP rollouts in the 1990s succeeded or failed. The bet of our AI program at OneDigital is that we have to treat it that way. As an HR transformation, not a technology rollout.

The methodology, in plain language

I run an AI Coworker program. Coworkers are conversational AI systems that have job descriptions, knowledge bases, named human supervisors, and probation cycles, and that anyone in the company can talk to through our internal tools. Some of them are practice-specific. Some are cross-practice. All of them go through the same lifecycle.

The lifecycle has three phases. Internship is where a small group of users stress-tests a brand-new Coworker against real work, surfaces the failure modes, and feeds the corrections back to a human supervisor. The Coworker is, by design, a rough draft at this phase. I tell every cohort the same thing: the AI in front of you is going to make mistakes; your job is to find them. That framing is the whole game. If users come in expecting a finished product, the program fails because the bar is set wrong. If they come in expecting an intern, the bar gets met.

Apprenticeship is the validation phase. The Coworker is opened to a wider but still curated audience for two or three weeks. The supervisor tightens the prompts, expands the knowledge base, retrains the system on the edges that came up in Internship.

Full-time is when the Coworker goes available to the whole company. Conversations are retained for context. Users can share threads with colleagues.

The structurally important detail, the one that does most of the actual lifting, is the supervisor. Every Coworker has a human supervisor, and that supervisor sits in the business unit, not in the tech organization. The supervisor is whoever is the best at the relevant job already. That choice is expensive. Pulling a top performer off active work to encode their craft into an AI is an opportunity cost the program owns directly. It is also the only way the Coworker becomes worth using. The Coworker is, at the end of Internship, a high-fidelity transmission of the supervisor's expertise. If the supervisor is mediocre, the Coworker will be mediocre.

This is the part nobody who hasn't done it gets to feel. The leverage point is the supervisor, not the model. The model is interchangeable. The supervisor is not.

What gets paved over in keynote decks

There are a few details about the actual day-to-day that almost never make it into the marketing version of an AI program. Some of them:

A meaningful share of Coworkers don't survive Internship. They get cut. They didn't perform. We may revisit them in six months when the underlying models improve. Right now they're on the cutting room floor. The internal joke in the AI Programs team is that I have fired enough Coworkers by now that when AI becomes sentient I will be first in its firing line. The joke is sticky because it is operationally true.

A non-trivial share of Coworkers that survived Internship are currently on Performance Improvement Plans. The Supervisor and I review their KPIs together; we agree on what better looks like; we set a timeline; we either retire the Coworker or we graduate it. None of this is hypothetical. None of it shows up in the brochure.

The single biggest adoption lever we have found has nothing to do with the AI. Practices that put their supervisors on a biweekly facilitation cadence (twenty to thirty minutes of group office hours where consultants bring their hard cases and the supervisor demos how the Coworker handles them) saw adoption move from around a third of the unit to about three quarters of the unit, inside a quarter. The Coworker did not change in any of those quarters. What changed was the conversation around it. Nothing in that sentence sounds like AI. It sounds like the same change-management discipline that worked for CRMs in 2014, and the same that didn't work for the companies that skipped the supervisor cadence.

I am about a year into knowing this, and I still have to remind myself of it weekly. The instinct is to chase a new model release. The discipline is to keep showing up to the supervisor cadence.

The next problem I do not have an answer to

For all the methodology, the next problem is the one I haven't solved. Right now most of how people inside OneDigital use AI Coworkers is transactional. Review this email. Summarize this document. Generate this proposal section. Those are real uses. They save real time. They are not the actual leverage AI offers.

The actual leverage is collaborative. The Coworker as a thinking partner you push back against, rather than a faster output machine. The Coworker that helps you decide what the strategy should be, not just helps you write it down once you've already decided. The training works for the transactional use case. The supervisor model works for it. The infrastructure works for it. Most people do not naturally cross over to collaborative use, even when they could, even when the Coworker would handle it well.

I do not have an answer for how to move a workforce from transactional to collaborative. It is the chapter we have not written. It is also, almost certainly, the next year of work.

Why I trust this is real

The most useful test, when you are inside a program and not sure whether you are seeing a real pattern or a flattering one, is whether outside experts have started studying it. They have. Shikhar Ghosh, a professor at Harvard Business School and one of their senior tech-entrepreneurship faculty, recently published a case study on what we did, called Building a Digital Workforce. There is a setup case before it covering the strategic decision that led to the program. Both are now in HBS's case collection. Vinay's book about the methodology comes out this summer, with the same set of mechanics described here at much greater depth. A joint Harvard, MIT, and UCLA research paper studying the same topic is in motion for later this year.

What I take from all that is not that I have figured out anything brilliant. It is that the boring middle is real enough to study. Operators get paid to make it work. Pundits get paid to predict the future. The future, the part I see from inside, shows up in the Supervisor's biweekly 1:1 with the Coworker. It does not show up in any keynote.

What this means if you are buying AI

If you are at an enterprise that is buying AI right now, my honest read is that picking the right model is not the part of the work that determines whether you succeed. The frontier models are converging on capability fast enough that the difference between buying GPT-5 versus Claude Opus 5 versus the next thing is going to be smaller than the difference between running a real change-management program around either one and not running one.

Write the job description for the Coworker you actually want to hire. Find the top performer in the relevant role inside your company and ask them to become its supervisor. Build the Internship as a real quality-assurance cycle, not as a beta test. Run the biweekly facilitations. Cut the Coworkers that aren't earning their seats. Resist the temptation to skip the supervisor cadence because users want access faster.

None of this is futuristic. None of it is hard to explain. It is, on the inside, deeply unsexy. It is also the reason a 6,000-person traditional consulting firm has reached 52% workforce-wide adoption in the twelve months its AI Coworker program has been live, while the median enterprise AI program is sitting at roughly four times less weekly active use.

I do not think this is permanent. The models will keep improving. The supervisor model will shift. The Coworker idea may evolve into something else within a few years. I really hope it stays useful for at least a little longer, because the people who actually run these programs are still mostly trying to figure out the basics, and the basics are not what the keynotes are selling.