The People function is being rebuilt. Good.
Boards now expect People leaders to speak finance, own AI governance, and connect workforce decisions to enterprise value. Most of the field is scrambling to catch up to that expectation. I’ve spent almost twenty years building toward it — not because I predicted the technology, but because I always thought the job was bigger than the field admitted. Here’s how I actually think about the work, and where each idea came from.
People as a business accelerator
Most companies still run People as a support function: process the requests, run the cycles, present a quarterly deck six weeks after the moment it could have mattered. I run it as an operating function — the same instrumentation, discipline, and accountability to outcomes as any revenue-producing part of the business. Org design, talent, and rewards are levers on revenue, valuation, and retention, and every talent decision should be defensible in front of a board. Run that way, People stops being overhead and becomes an accelerant.
The test is simple: if your people strategy can’t survive contact with a CFO, it isn’t a strategy.
Framework 01The cheesecloth
Every workflow in my function passes through one question before anything else: why can’t this be AI? I call it the cheesecloth. What passes through is base-layer work — it should be automated, and if it isn’t yet, that’s the first problem to solve. What doesn’t pass through — the judgment calls, the trust-building, the decisions that touch a person’s livelihood — is the human work, and now it’s visible and named and you know what it’s worth.
The failure mode is layering AI onto a broken process. That doesn’t fix the process. It industrializes the dysfunction. The tool was never the strategy; the work redesign is the strategy.
Framework 02Prove the negative
Flip the default. The working assumption is that everything can be done by AI — then prove the negative. Human involvement carries the burden of proof, the same way headcount always has. You’ve always had to rationalize a hire, a promotion, a new layer; now you rationalize the human in the loop with the same rigor: why does this step need a person, and what specifically does the person add?
This isn’t robots running everything. There’s plenty AI can’t do. But organizations that start from “could AI do this?” quietly answer no and hire another person. Organizations that start from the reverse hypothesis find out what’s actually true.
Framework 03The 70% is the whole game
There’s a useful rule of thumb on where AI value comes from: roughly ten percent from the algorithms, twenty percent from the data and infrastructure, and seventy percent from the people, process, and organizational change around it. The field spent a year worrying AI would shrink the People function. It did the opposite. It handed us the seventy percent — the part that actually determines whether any of it works.
Framework 04The third column
The org chart had two columns: managers and individual contributors. It now has a third — agents, in real boxes, with charters, owners, and performance expectations. I watched this land for a leader who ran a small team and worried his people had nowhere to grow. If AI does the work, where are the promotions? There are plenty, I told him — they’re just not management of people. You can be an individual contributor with six direct reports, and all six are agents. The manager doesn’t disappear. What they manage changes.
Three things follow. Workforce planning becomes a statement of combined human-and-agent capacity, not a headcount number. Learning discovers its fourth audience — after employees, managers, and executives, you’re now teaching the models themselves: context, standards, judgment. And governance stops being a policy document and becomes a bright operating line.
Who owns the consequence
Half of business decisions will be AI-augmented or automated before 2027. That number has stopped being controversial. The question underneath it hasn’t been answered fast enough: when a black box makes the call and gets it wrong, who owns the consequence? People feel less ownership of a decision a machine made, and they get looser with the truth when they can hand it off. So in every system I build, a human owns every employee-affecting outcome — by name, in the moments that matter. Not because AI can’t decide. Because someone has to be accountable when it does.
Framework 06Impact over performance
A decade ago, as CPO at Altisource, we killed the annual performance review and replaced it with the Impact Review. Words matter because mindset matters: the question stopped being what people were doing and became what their work changed — quantifiable business impact. If you ran a twenty-person team and delivered X, and someone else ran a two-person team and delivered 2X, the second person was more valuable. Levels, compensation, everything followed from that logic. It was bleeding-edge then. Today “impact” shows up in company values everywhere — and AI is about to finish the argument, because when agents absorb the activity, activity stops being worth measuring. Impact is all that’s left.
Unmeasured talent decisions accumulate like unpaid interest — I call it talent debt, and most organizations are carrying more of it than their balance sheet will ever show.
Framework 07Watch the tape
Years ago I trained toward — and was invited to try out for — NBA officiating. The craft behind that invitation taught me more about development than any HR certification ever has. Refereeing is mostly not the game — it’s the film. Three to four hours reviewing a one-hour game, grading hundreds of decisions on a spreadsheet, including the calls you didn’t make. Clipping your worst moments and sending them to people better than you for brutally open feedback. Like an iceberg, ninety-five percent of the work is never seen.
Now hold that up against how the average executive develops. Almost every meaningful interaction a modern leader has is on Zoom or Meet or Teams. That’s the game tape — recorded, transcribed, sitting on a server doing nothing, while leaders operate on memory and self-perception. AI just made the film room free: a private after-action report on every meeting you run. You spoke seventy-eight percent of the time. Three of your four questions were rhetorical. The person who hadn’t spoken still didn’t. Coaching doesn’t get replaced by this — it gets amplified. The leaders who build this system in the next couple of years will compound an advantage the rest will struggle to close.
Framework 08The tells
I can usually tell within minutes whether a team has actually adopted AI or is performing adoption. Three tells. They share prompts but never what they built with them — the focus is on the input, not the output. AI comes up as something they attended a training on, not something they shipped work with. And the one that settles it: a manager can describe their team’s AI use in detail but goes quiet on their own. That last one is the whole thing. The personal commitment can’t be outsourced. Where the leader hasn’t made it, the team is decorating workflows, not redesigning them.
The closeThe human side is the hard side
The AI moment has done something I didn’t expect: it’s made the best leaders I know more human, not less. When AI absorbs the work that used to fill the hours, what’s left is the part that can’t be automated — the relationship, the judgment, the hard conversation, the presence. I trained in theater before I trained in HR, and it shows up everywhere: culture is performed, not posted; feedback is a craft, practiced like a foul call in a mirror until it looks effortless.
The human edge isn’t a capability you train into an organization. It’s a decision leaders make, every day, about what they refuse to delegate — not because they can’t, but because the act of doing it is the value. Build the systems. Run the workflows. Embed the agents. Then make sure you haven’t optimized your way out of the reason any of it was worth doing.
Philosophy is cheap. The record is on the next page →