"You may only use Copilot."
And everyone has ChatGPT open anyway.
Welcome to FullStack HR, and an extra welcome to the 102 people who have signed up since last week.
If you haven’t yet subscribed, join the 11500+ smart, curious HR folks by subscribing here:
Understanding Claude’s potential and limits in HR.
On 15 July I’m joining Kinfolk for a conversation about the current stage of AI Adoption and where teams are building internally vs evaluating external solutions at the moment.
Most HR and People Operations leaders I work with are deep into the pilot stage now. The enthusiasm is real but many organisations are still working on what a sustainable AI setup looks like for an HR team, outside of individual productivity.
Building with Claude for individual work makes sense. Drafting policies, summarising documents, writing job descriptions. But when teams try to extend that into their operational layer, and start working on the employee lifecycle requests, the data flows, the governance, the accountability, the gap between “this works for me” and “this works for my team at scale” becomes clear very fast.
We’ll be working through the build vs. buy decision with the infrastructure context it deserves and understanding what are good use cases for internal builds with tools like Claude (or Copilot) and where it’s worth exploring a third-party solution.
Wednesday 15 July · 5pm BST / 9am PT / 12pm ET
Happy Tuesday,
Listen to this one on Spotify.
Last week I was in Almedalen.
For those of you outside Sweden, Almedalen is our big political week. Once a year, the whole country more or less relocates to Visby, a small medieval town on Gotland, and for a week, politicians, business leaders, lobbyists, journalists, and everyone with an agenda gather in the same few streets to talk. It is where politics and business meet. It is loud, it is intense, and this year there was a lot of talk about AI. (Surprise, surprise.)
And somewhere in the middle of one of these sessions, it hit me. All these politicians and business leaders, the people shaping how the country thinks about this technology, their version of AI is mainly…Copilot. (Or at best the free version of ChatGPT.)
But that’s their reference point. That is the picture in their heads when they stand on a stage and talk about what AI can and cannot do for society. And I could not stop thinking about it for the rest of the week.
I think this is because almost nobody picks their own model. People ride along with whatever their organization landed on, and eight times out of ten that means Copilot.
And the way it got chosen is usually no choice at all. It got chosen on the logic of “well, we have Microsoft, so this is the model we have.” Congrats, good luck, and you, dear colleague, get to adapt to it. Nobody ran a needs analysis and nobody asked the one question that matters: Does this solve the right problem for the work we do?
Don’t get me wrong. Copilot has gone from a so-so pick to a fairly okay one. It is not a bad choice. You can defend it. But is it the best and only model for an organization? There my answer is, without hesitation, no.
So let me explain what I mean, because I don’t want this to read as brand snobbery.
We already accept that different jobs need different tools. We pick different computers depending on the work. We pick different software depending on the craft. For some people, Photoshop is mission-critical. For others, it is completely irrelevant. Nobody blinks at that.
So why do we stop thinking the second we get to AI?
Because when we stop thinking and just default to Copilot, two things break. And both are quietly expensive.
You build the wrong picture of what is possible
The first thing that breaks is your sense of what AI can even do.
Organizations build their entire expectation of what is possible on the one model they have touched. To be precise, it is the people inside the organization forming that impression. But the effect is organizational. The shared sense of “this is what AI is” gets anchored to a tool that is not at the front of the field.
Think of it like a screwdriver. Before you had a proper screwdriver, getting a screw out was hard, so of course the screwdriver felt like a leap. It works. It makes the job more efficient. And if a screwdriver is all you have ever held, you genuinely believe you have arrived at the utopia of screwing in a screw. What you don’t know is that there is a power drill sitting on the shelf, basically frictionless, that does the whole job while you press one button. That is roughly where we are with these models. What you were handed is the screwdriver. Claude and ChatGPT are far closer to the power drill.
And when the screwdriver is all you have, you set the wrong expectation for the entire organization. No wonder people feel lost on AI when their whole worldview is built around that one tool and nothing past it. Which, to bring it back to Almedalen, is exactly the level the conversation was stuck at.
There is a timing piece here too. Technically, raw problem-solving stopped being the bottleneck for most organizations at the start of this year. From the arrival of Opus 4.5, if you picked the right model and gave it the right conditions, the problem-solving most organizations need has been well within reach. But if everything you have ever seen is the default tool, you missed that shift completely. And then I understand why you are still hiring for problems that are already solved.
Your best people are already working somewhere else
The second thing that breaks is shadow IT, and it is still everywhere.
I have lost count of how many people I have met who got hard, explicit instructions. You may only use Copilot.
And then every single one of them has ChatGPT open on the side. The people who want to be further ahead, the ones pushing this forward, the internal enthusiasts you most want to capture and put to work, those are exactly the people who quietly switched to something else.
That is not good from a lot of angles. Not good for security. Not good for you as an employer, because the data is going somewhere you have zero visibility into. There is a whole pile of problems, but security and knowledge sharing are almost the first to go. When people work in the shadows they tell you less about what they did, and the openness to share evaporates. You lose the exact thing you wanted, which is people building out in the open, where you can see it and learn from it.
And soon this becomes an employer branding question
The third is that over time I believe this becomes an employer branding issue.
Think back to how it went when the phone arrived, or when social media arrived. (For those of us who were around for it.) Not being allowed to use LinkedIn at work used to be a real thing. People argued about it. Today it is a non-issue at almost every workplace. In a lot of places you are expected to share what you do on LinkedIn. The same noise happened every time one of these things first showed up, and then it faded.
The same arc is coming here. You will have someone arriving fresh from university, or from another company, used to working with their own agents, used to bringing their own reality and their own setup with them. And then they walk into a locked-down corporate environment and meet nothing but flatness compared to what they had in Codex or Claude Cowork. I know this happens, because I have met plenty of people living that exact gap, people who keep actively working in something else on the side. We can argue about the scale. But I know it is still going on. So it becomes a question candidates start asking. Which model do I get to use? What does my token budget look like? How am I allowed to use all of these?
So what do you do about it
You secure all these models and offer employees the choice. Choose what you prefer working with, but in our secure environment.
I know that is a cost, but so is a data breach, and so is the cost of people working on their own, without sharing the benefits of what they do.
So in most cases, my question is simply this. Why not? What is the reason, other than cost, for not letting people choose between Claude, ChatGPT, Copilot, and Gemini? You can offer all four.
So what is the reason you are not doing it?
If you tell me it is security, compliance, governance, then what you have is a lazy IT department that cannot be bothered to do the work. That is it. There is nothing else. Make no mistake, all of this can be made safer, but it takes effort.
This is going to become an employer branding question and an organizational question, and the organizations that solve it fastest are the ones that win.