Most companies are making the AI decision backward. They buy the agents and the applications, push them onto existing roles, and hope the strategic outcomes follow.
That is the inversion. It lets the tool write the strategy. And it is exactly why the returns keep not showing up.
The right order runs the other way. You name the outcome first. Then you point AI at the levers most able to move it. Only then do you redesign the roles. This is role re-engineering, and it starts with strategy, not software.
The inversion is happening even at large companies
This is not a small-company mistake. It is happening inside big, sophisticated organizations too.
The evidence is clear. In one study of large-company CFOs, only 12% felt very prepared to manage the workforce shift AI creates. Meanwhile, those same leaders buy AI tools faster than they adjust their talent strategy.
So the tools arrive first. The role design comes later, if at all. That is the inversion in one sentence. And the cost is not small. When the tool leads, no one rethinks the roles. So the work changes, but the jobs stay frozen in place.
The pressure underneath it is real, too. By 2030, employers expect 39% of core skills to change. IBM found the same convergence from another angle. In its 2026 study, 77% of leaders say talent and technology roles are now merging.
So the work is shifting fast. The question is whether you shape that shift or let the software shape it for you.
Start at the top, not at the role
Role re-engineering does not start at the role. Instead, it starts two levels up, in the C-suite.
Here is the sequence. First, leadership names the outcome that matters most. Then it identifies the highest-impact levers on that outcome. Next, it decides where AI can move those specific levers. Only then does role design begin.
So what is a lever, concretely? A lever is a specific driver of a specific outcome. In a services firm, it might be speed-to-proposal. In a product business, it might be margin on one line. It is never “efficiency” in the abstract.
The work, then, is to find the few levers that matter and aim AI directly at them. After that, you redesign the roles around those levers. And you upskill people to own them.
Take speed-to-proposal as the lever. The outcome is a faster path to more closed deals. So you point AI at the drafting, the research, and the pricing model. Then you redesign the seller’s role around judgment and the client relationship. The tool serves the seller, not the reverse. Role re-engineering means aiming the role at a named outcome.
Two companies, one tool, opposite logic
The contrast is easiest to see in two real cases. The difference was not the technology. Both companies bought capable tools. The difference was what each one decided the tool was for.
McDonald’s aimed AI straight at a role. It spent three years testing a voice bot to run the drive-thru order-taker job. But the real environment demanded judgment the tool could not deliver. Orders went wrong, and in 2024 the company pulled the pilot. The tool was the goal, so the goal became the problem. The lesson was not that AI failed. The order did.
Johnson & Johnson ran the opposite play. First, it let roughly 900 AI experiments bloom across the company. Then it measured the results, and the pattern was stark. Only 10 to 15% of those use cases drove 80% of the value.
So J&J stopped scattering. It concentrated on the few levers that mattered most. And it redesigned the work around them, including a copilot that makes its sales role sharper, not just faster. That is the move most companies skip. They add tools. J&J added outcomes.
This is role re-engineering, done right
J&J did not ask how many widgets it could push into a job. Instead, it asked which outcomes deserved the effort.
That is the whole discipline. First you name the outcome and the lever. Then you redesign the role and upskill the person to reach it. I wrote about why that human layer holds its value in *The Price Just Went Up on Humanity*.
McDonald’s optimized for the tool. J&J optimized for the outcome. Same technology, opposite results.
What this means for real leaders in plain English
Do not start your next AI project by asking which roles can absorb a tool. Instead, start in the C-suite with a sharper question.
So ask what outcome matters most this year. Then ask which few levers move it. Then ask where AI can move those levers hardest. Only after that do you touch the roles.
Role re-engineering is a top-down act. It cascades from strategy down into the job. It never pushes up from the tool. The order is the strategy. Get it right, and AI compounds. Get it backward, and it drifts.
Where Alchemy comes in
At inflection points, the difference between an AI experiment and a real return is the order of the decisions.
Alchemy runs role re-engineering from the top down. We help you name the outcomes, find the levers, and redesign roles and people around them. Then AI serves the strategy instead of setting it.
If you’re ready for a conversation: https://calendly.com/brynn-alchemyleadershiplab
References
- World Economic Forum, Future of Jobs Report 2025 — https://www.weforum.org/publications/the-future-of-jobs-report-2025/
- PYMNTS Intelligence, “No Roadmap, No Problem: How Enterprises Are Reinventing the AI Workforce” (2026) — https://www.pymnts.com/news/artificial-intelligence/2026/ikea-turned-8500-call-agents-into-design-consultants/
- IBM, 2026 CEO Study — https://www.ibm.com/thought-leadership/institute-business-value/en-us/report/2026-ceo
- PYMNTS, “Johnson & Johnson: 15% of AI Use Cases Deliver 80% of Value” (Apr 2025, reporting WSJ) — https://www.pymnts.com/news/artificial-intelligence/2025/johnson-15percent-ai-use-cases-deliver-80percent-value/
- CNBC, “McDonald’s to end AI drive-thru test with IBM” (Jun 2024) — https://www.cnbc.com/2024/06/17/mcdonalds-to-end-ibm-ai-drive-thru-test.html
- Alchemy Leadership Lab, “The Price Just Went Up on Humanity” — https://alchemyleadershiplab.com/price-just-went-up-on-humanity/
- Alchemy Leadership Lab, “The Technology Lean Out” — [confirm slug in WordPress]
Estimated reading time: 5 minutes