Insights
Human-Centered AI, and what the evidence says.
My working philosophy on enterprise AI — and the research that keeps arriving at the same conclusion. Every figure below links to its source.
I believe AI should amplify human capability — not simply automate tasks.
That can sound like a soft position. It isn’t. It is the harder one to execute, and it is increasingly where the money actually goes.
The automation-first instinct is the intuitive one: find the tasks a machine can do, remove them, count the savings. The scale of what is now technically possible makes that instinct feel urgent. But the organizations getting real value are spending their budgets somewhere else entirely.
spent on capability building and adoption for every $1 invested in
agentic technology — and $3 on process redesign.
McKinsey, August 2026
of companies report meaningful financial impact from AI — fewer
than. Ninety percent are investing.
McKinsey
Global Institute, January 2026
of the human skills employers seek today will endure, even as up to 57%
of US work hours become automatable.
McKinsey, August 2026
of leaders report making progress designing effective human-AI
interactions.
Deloitte, March 2026
That first number is the one I would put in front of any executive committee. McKinsey found that for every dollar organizations invest in agentic technology, they spend three on process redesign and five on capability building and adoption.
One to three to five. The technology is the smallest line item.
Which means enablement is not the soft wrapper around the real work. It is the majority of the work, and budgeting it as an afterthought is how programs quietly fail.
The failure mode shows up clearly in the numbers. McKinsey Global Institute reports that ninety percent of companies are investing in AI while fewer than forty percent see meaningful financial impact. Their diagnosis is blunt: “Layering a chatbot or automation tool onto those legacy processes yields incremental gains at best.”
Task-level automation does not get you there. Workflow redesign does. That distinction is the whole job.
It also changes what we should expect to happen to people. Even with up to 57 percent of US work hours technically automatable, more than 70 percent of the human skills employers seek today are expected to endure. McKinsey describes the work that emerges around AI — guiding, prompting, validating, refining and building on what the system produces — alongside a rising premium on critical thinking, judgment, coordination and the social and emotional skills that don’t substitute cleanly.
Skills are being recombined, not erased. Which is why I think the more useful question is not what AI will take over, but what it frees people to become exceptionally good at next.
The distance between believing that and doing it is enormous, and it is where I spend my time. Deloitte’s 2026 Global Human Capital Trends found that 85 percent of leaders consider adaptability critical — while only 7 percent say they are leading at helping their workforce grow and adapt, and just 6 percent report progress on designing effective human-AI interactions.
Read that last number again. Ninety-four percent of organizations have not yet designed how their people and their machines actually work together. That is not a technology problem.
What closes the gap is unglamorous and human. The same McKinsey work found that people who distrust their organization’s support during an AI transformation are 1.5 times more likely to feel anxious about it, and that high performers are three times more likely to say senior leaders visibly own the effort. Trust and leadership attention are not soft accompaniments to the rollout. They are load-bearing.
So the philosophy comes down to three things I try to hold together:
The working definition
Human-Centered AI as the design principle — amplify capability, don't just remove tasks. Responsible AI as the discipline that keeps it trustworthy inside a regulated institution. Human potential as the actual point of the exercise.
Sources & further reading
What I'm reading on this
- Agentic AI change management: Closing the adoption gap Where the money actually goes: one dollar of agentic technology against three of process redesign and five of capability building. Also the trust and leadership findings that separate high performers from everyone else.
- The rise of the human–AI workforce The work that emerges around AI rather than disappearing beneath it — guiding, validating and building on what the system produces — and the rising premium on judgment and coordination.
- A new year’s resolution for leaders: Redesign work for people and AI Ninety percent of companies investing, fewer than forty percent seeing financial impact, and the reason why: layering tools onto legacy processes yields incremental gains at best.
- From tensions to tipping points: Choosing the human advantage The gap between what leaders say about adaptability and what their organizations have actually designed — including how few report progress on human-AI interaction.
Figures cited as published; linked to the original research so you can check my reading of it.
Comparing notes
The technology arrives on its own.
The capability doesn't.
If you're working on AI enablement somewhere — or just thinking about the same problems — I'm always glad to trade notes on what's actually working.