For twenty years, "enterprise software" meant one thing: tools that wait.

They waited for you to click. Waited for you to upload. Waited for you to prompt. The smarter they got, the more elegant the waiting. That is what Copilot is. That is what ChatGPT-in-a-browser is. That is what every RPA platform your CIO bought in 2019 is, once you strip the marketing.

This is not a semantic shift. It changes what you buy, what you measure, and what your org chart looks like in 2028. If you are a CEO or founder still evaluating AI as a "tool category," you are optimizing for the last war.

Click to zoom

Why "copilot" was always the wrong word

A copilot sits next to the pilot and helps them fly. That is precisely the wrong metaphor for what is useful now.

You do not need help writing one email faster. You need the ten emails written, the three documents reviewed, the contract redlined against your playbook, the pipeline updated, the customer's question answered at 2am, and a summary waiting for you at 8am — while you work on the one thing only you can do.

The copilot paradigm caps value at your attention. Your attention is the bottleneck the whole category was supposed to break.

Digital coworkers break it. They do not assist your work. They do the work.

Click to zoom

First principles: what actually makes something a coworker

Forget the marketing. Here is the test.

These are not aspirational. They are the line between the category that is dying and the one that is emerging. Every enterprise AI decision you make in the next three years is, underneath the features and the pricing, a decision about whether you believe in tools or in coworkers.

The numbers are visible. The value is not — yet.

The adoption data tells an uncomfortable story. According to McKinsey's 2025 State of AI¹, 88% of companies now use AI in at least one function. But only 23% have scaled it across the enterprise. And only 6% are capturing what McKinsey calls "disproportionate value" from it.

Click to zoom

That gap — between broad adoption and concentrated value — is not an accident. It is what happens when 88% of companies buy tools and 6% build organizations around coworkers.

The economics reinforce the gap. A human customer-service interaction costs between $2.70 and $5.60. A well-deployed AI coworker costs $0.50 to $0.70 — and is online at 3am on a Sunday. A preregistered BCG–Harvard study of 758 knowledge workers² found AI assistance made them 25% faster and produced 40% higher-quality output on tasks within the model's frontier. GitHub's enterprise data³ puts the coding speedup at 55%.

None of this is news. What is news is that the curve is not linear. Jensen Huang keeps saying "100 AI workers for every human" like he is being provocative. He is being conservative. Once you can manage one coworker reliably, you can manage a hundred. Once you can manage a hundred, the constraint on your company stops being talent acquisition and starts being taste.

What actually changes for a CEO

If you accept the shift, four things change. None of them are about "AI strategy."

1. 01

   You stop buying software. You start hiring.

   A coworker does not get rolled out by IT in Q3. It gets onboarded — given a role, a scope, access, and a manager who is accountable for its work. That is a hiring process, not a procurement process. Your procurement team is not ready for this. You will have to make them ready.

2. 02

   You stop reorganizing IT. You rewrite the org chart.

   The design question of the next five years is not "how many people do we need?" It is "what is the right ratio of humans to coworkers in each function, and who manages whom?" Legal ops with three lawyers and twelve coworkers looks nothing like legal ops with fifteen lawyers.

3. 03

   You stop measuring adoption. You measure output.

   "How many seats are active?" is a tool metric. The right metric is the same one you would use for any other employee: how much work got done, at what quality, at what cost. If your internal AI reporting is a usage dashboard, you are still thinking in tools.

4. 04

   You stop training people on software. You start training coworkers on your business.

   Institutional context — the unwritten rules, the history of why you do things the way you do them — is the hardest asset to transfer to any new hire. A coworker that learns it once and remembers forever is a fundamentally different kind of asset than a tool every new hire has to rediscover.

The skeptic's case, honestly

The counterargument is real. Today's agents still fail. They hallucinate. They get tricked by adversarial inputs. Reliability is nowhere near 100%.

You do not fire a new hire because they need supervision. You scope their work, give them the systems they need, check their output, and expand their role as they earn it. The same management discipline applies here — and it is why "digital coworker" is a more honest metaphor than "autonomous AI." Accountability is not a bug; it is the whole point. A coworker is something you are responsible for. That is precisely what makes it ready for regulated work, where "the AI did it" is not a legal defense.

The companies that will win this decade are not the ones with the best models. They will be the ones with the best management discipline for coworkers they do not have to pay a salary to.

The only question that matters

In five years, every company will have digital coworkers. That is no longer a forecast.

The open question is which companies will have designed themselves around coworkers, and which will be pasting outputs from tools into spreadsheets while a smaller, faster competitor eats their lunch.

The 6% capturing disproportionate value today are not the ones with the biggest AI budgets. They are the ones who stopped thinking of this as software.