The Sand Trap
Why Your AI Strategy Is Optimizing the Wrong Layer
A manager spends twenty minutes prompting an AI to rewrite a Slack message that would have taken ninety seconds to type. He posts about it as a productivity win. Most of HR’s AI adoption right now looks exactly like this. We’re using a sledgehammer to adjust a picture frame, and calling it transformation.
Think about your day as a jar. Big rocks first, the strategic work that defines your role. Medium rocks next, important but not foundational. Then pebbles and sand, the small administrative tasks that fill whatever space is left.
Most AI adoption in HR right now is happening entirely at the sand level. Drafting emails. Summarizing meeting notes. Cleaning up a job description. Useful, sure. But sand was never the problem. Sand fits itself into whatever space remains, that’s the whole point of sand. You could automate every grain of it and your week would look basically the same, just with more open calendar blocks you’ll fill with more sand.
The rocks are where AI could actually change something, and almost nobody is using it there. I built a conversational AI agent that conducts talent assessments for managers, something that used to require a facilitator, a calendar, and ninety minutes per person. It now runs without me, at scale, for over 250 managers. That’s not a sand-level task. That’s a structural shift in how an entire function operates.
The assessment process didn’t get faster. It got rebuilt. The difference between using AI to write faster and using AI to rethink how the work gets done in the first place is the difference between efficiency and transformation, and most HR leaders are still stuck on efficiency.
The rock-level work is harder to hand to AI because it requires us to actually examine what we do and why, not just how fast we do it. Sand-level prompting doesn’t ask you to confront anything about your function’s relevance. You can prompt your way through a hundred emails and never once ask whether the process generating those emails should exist.
Rock-level thinking does ask that. It forces the question of whether the thing you’ve always done is still the right thing to do, and that’s a much harder conversation than “can you make this sound more professional.”
This is also why so many AI rollouts stall at the sand layer and never move. It’s not a skills problem. It’s that rock-level redesign requires authority, time, and a willingness to be wrong about how your function has always worked, three things that are in short supply for most HR practitioners.
If AI can already do your sand, what does that free your rocks to become, and have you actually let it change anything yet?