AI Is a Work Tool, Not a Perk
Stop rewarding curiosity with access and start treating AI like email
Welcome to FullStack HR, and an extra welcome to the 13 people who have signed up since last week.
If you haven’t yet subscribed, join the 11200+ smart, curious HR folks by subscribing here:
Cowork cohort Beta 2 is starting soon!
The early bird price is gone, but there is still room for a couple more people.
It is very practical, and even though we talk about Claude Cowork, you can use the same principles for Copilot Cowork or OpenAI’s Codex.
Sign up here.
Happy Tuesday,
101, that’s how many lectures / workshops / sessions or call it whatever this year.
I know it’s too much, but as said before, the true upside of this is information back to me. You see what’s working in organizations and what’s not.
Every client's work is a data point if treated as such, which builds a clearer and clearer picture of how to do this whole AI transformation. Most of my articles these days stem from that collective knowledge. This one certainly does.
So let’s get to it.
There have always been fast adopters and slower adopters in workplaces. The volunteers and the volun-tolds.
Most of the time, that gap closes on its own. Someone shows the slower group a shortcut, a system gets rolled out company-wide, training fills in the rest.
Generative AI is different, and what many employers are doing right now might risk making the gap permanent.
A group of employees, often a small one, has figured out generative AI on their own. They started with the free version of ChatGPT at home. They got curious and now use it daily.
Another group, often a larger one, has not done this. They have not been pushed by curiosity, they have not had time, and they have not seen anyone they trust show them why it matters.
People adopt new things at different speeds, so far, nothing new there.
But, and this is the important thing here, what a lot of employers are doing right now is to look at the situation and say, sensibly enough, “let us give the people who are already using AI the proper tools.” Copilot for Microsoft 365 Business, ChatGPT Enterprise or Claude Enterprise. The full model with the safety guardrails, the data protection, and the integrations.
Then the employer looks at everyone else and says, “and you get Copilot Chat. Or nothing. Because it costs too much to give the full thing to everyone.”
So we are giving the most powerful productivity tool of our generation to the people who are already pulling ahead. And to the people who were already behind, we are saying, “here, take this stripped-down version. Welcome to your paper and pencil.”
And in one stroke, we’ve widened the gap even further between the two.
This is a work tool.
Generative AI is not a perk for the AI-curious. It is not a power-user feature for the technically inclined. It is a work tool. The same way a laptop is a work tool, the same way email is a work tool.
You do not give email only to the employees who showed early interest in email. You give it to everyone because the work requires it. We are at the same point with AI. The work requires it, so everyone needs it.
The fact that some employees got there first on their own does not change the answer. It only tells you who needs less training, it does not tell you who deserves access.
I have been saying this for years. I am sorry, IT, for repeating myself. But we have to start working together on this one. In most organizations the AI tooling decision lands in IT. Which makes sense on the surface. IT handles licenses, security, and deployment.
But when IT owns the rollout without HR and leadership in the room, the framing shifts. The tool becomes something for technical people and early adopters who knock on the door. The conversation turns into “who has shown they can use this responsibly,” “who has the use case to justify a license,” “who can articulate the ROI.”
That sounds reasonable, but it is also exactly how you end up with two-tier deployment.
IT is not the enemy in this story. IT is doing what IT has always done, which is roll out technology safely, with use cases and license discipline. That is the right behavior for ninety percent of what IT deploys.
This is the ten percent where it is the wrong behavior. Because this is not really a technical rollout. (Once again, repeating myself here but still…) It is a change in how people work, and how people work belongs to HR and to operational leadership, not to a procurement spreadsheet.
The organizations that get this right have leaders who walked into the room and said something different. This is not a technical tool. It is a work tool. Everyone gets it. We train everyone, we support everyone, we expect everyone to figure out their own use cases.
The best leaders I have worked with do three things at once.
They redraw the map. AI access is decoupled from technical fluency. It sits where laptops and email sit. Default on for everyone.
They invest in showing people what it is for. Not a generic intro session at lunch. Concrete, role-specific examples from peers. The communicator sees what another communicator did with it. The HR partner sees what another HR partner did. The procurement specialist sees what another procurement specialist did.
They protect the time it takes to learn. They make it explicit that the first few weeks of regular use will feel slower, not faster, and that this is fine. They build in space for the awkward middle phase that most adoption efforts skip.
Here’s your fax machine.
Imagine it is 1995 and your company is rolling out personal computers.
You give the desktop PC to the employees who already use a computer at home in the evenings. Same applies to anyone who has shown technical curiosity.
Everyone else gets paper and pencil. “Sorry, we noticed you have not shown enough interest in computing. Maybe next year, if you can demonstrate some initiative.”
Nobody would do this. The absurdity is obvious.
Yet that is what we are doing with generative AI in 2026. The only difference is that the gap is fresh enough that the consequences have not yet hit the labor law system or the union negotiation table.
They will.
I am not a lawyer. But systematic differentiation in access to fundamental work tools, based on something as vague as “engagement” or “self-driven curiosity,” is going to be challenged. When AI access becomes a measurable factor in performance, productivity, and promotion, and it already is, then unequal access becomes an HR risk.
Imagine it’s 2028. A union representative looks at your AI tooling policy and asks, “why did you give the Enterprise license to person A and the basic version to person B, when they are in the same role?”
“Because person A used it more on their own time” is not going to be a defensible answer.
It will look like one of those decisions everyone wishes they had pushed back on at the time.
What good looks like
Either your organization is going to work with generative AI as a core capability, in which case everyone gets it. Same access, same model, same support. You differentiate by training intensity and use case, not by license tier.
Or your organization is not yet ready to deploy at scale, in which case nobody gets it yet. You run controlled pilots with clearly defined participants and timelines. You learn, you decide. You roll out to everyone, or you do not roll out at all.
What you do not do is the middle path. The “let us give it to the keen ones and see what happens” path. It looks pragmatic in the short term and looks like discriminatory tooling in the long term.
I am pushing it hard here. The middle path is where most organizations sit right now, often by accident more than by design. That is the pattern that leads nowhere.
What to do.
If you are an HR leader or a CIO reading this and recognizing your own organization, here is the move.
Pull the data on who currently has AI licenses. Look at the distribution by role, level, location, gender, age, and tenure. Be honest about the patterns you see. Most of the time, they are not pretty.
Decide. Are you in or are you out? Write it down. Get the executive team to sign it.
If you are in, build a deployment plan that gets the full tool into every role within twelve months, with training that meets people where they are.
If you are not in yet, stop the partial deployment. Pause the licenses. Run a real pilot with clear boundaries and a real evaluation.
Tell your employees what you decided and why. The worst thing you can do is leave the current quiet two-tier system in place and hope nobody notices.
We talk a lot about AI replacing jobs. (I did it last week.) The more pressing question, for the next few years, is which humans get to use AI and which humans have to compete against the ones who do.
That decision is being made right now in procurement meetings and IT budget reviews. Often without HR in the room, without a labor lawyer in the room and without any reflection on what kind of workplace this creates.
Then in two years we will wonder why engagement scores diverged so sharply between departments. Why the same role has wildly different output. Why people in their fifties are quietly being managed out. Why the union is suddenly very interested in tooling policies.
It will not be a mystery, we will have done it to ourselves.
The fix is not complicated.
Treat AI as a work tool. Give it to everyone, train everyone, and support everyone. Redraw the map so AI access is not coupled to who happened to show technical curiosity first. And do it with IT, not around IT.
Anything less and you are building an A-team and a B-team inside your own walls.
Once that gap is locked in, it is much harder to close than to prevent.
I reckon employees are going to eventually be picky about the tools they expect - my ideal stack would be Macbook + Gsuite + Claude.
I would struggle to move to a Windows + Outlook + Copilot ;-)