Gen AI & Agentic AI: Reimagining Workflows
The capability the 2023 book didn't have. Generative and agentic AI don't pay off when you bolt a copilot onto the old way of working — they pay off when you redesign the workflow around them.
What changed›
- New chapter — added in the June 2026 restructure.
This is the capability the first edition didn't have a chapter for. The lesson of the agentic years is sharp: the value isn't in adding AI to the old workflow. It's in redesigning the workflow around it.
When generative AI arrived, the instinct everywhere was to bolt a copilot onto existing work — a chat box beside the same process, helping each person do the same job a little faster. It scaled beautifully and moved almost no money. That's the central finding of the agentic era, and it's why the second edition of Rewired added a whole chapter on reimagining workflows: the returns don't come from assisting the old way, they come from rebuilding the way itself.
Copilot bolted on — what fails
Same workflow, plus an assistant. Horizontal, easy to roll out to everyone — and it barely moves the number, because the process around it never changed.
Workflow rewired — what good looks like
One process redesigned end-to-end around agents doing real steps. Vertical, harder to build — and where the economic value actually sits.
The horizontal-versus-vertical split is the heart of it. Broad copilots spread fast and capture little; vertical use cases — agents wired deep into one specific workflow, doing steps rather than chatting — carry the value but are far harder to scale. That difficulty is exactly why so many efforts stall: by most counts, only a small minority of agent pilots — well under a quarter — ever reach real production. And the companies that break through share a tell: the high performers are roughly three times more likely to have fundamentally redesigned workflows rather than layered AI on top of them.
Agents are workflow automation, not chat
The mental shift this capability demands is to stop thinking of AI as a smarter assistant and start thinking of it as a new kind of worker in the process — one that takes actions, with humans setting direction and owning judgement. (The Automation & Agents course is the engineering of this; here it's the strategy.) Which work goes first is a prioritisation call, and it's the same value logic as the roadmap: pick the workflow where a redesign captures a real, sized prize, not the one that demos best. Reimagining everything at once fails; reimagining the right one domain, deeply, is how this capability actually pays.
Where it goes wrong
Buying everyone a copilot, calling it an AI transformation, and waiting for earnings to move. They won't — you've made the old workflow slightly faster, not built a new one. The opposite miss is trying to rewire every process at once; pick one high-value workflow and redesign it deeply before you spread.
Try this
Take a workflow where you're “using AI” today and ask the blunt question: did the process change, or did we just add an assistant beside the old one? If it's the second, you've found why the value isn't showing up. Then pick the single workflow where a genuine redesign — agents doing steps, the process rebuilt around them — would capture the biggest sized prize, and start there.
Grounded in Rewired 2nd ed. (McKinsey, 2025), “Reimagining Workflows with Agentic AI,” and McKinsey's Seizing the agentic AI advantage and The state of AI.