IKEA's innovation team asked AI to design a couch that broke every rule. It gave them a sofa: boxy, cushioned, the average of every couch it had ever seen.
So they put the laptop away and talked about campfires, about people gathering and about what a couch is for. Paula Goldman, who tells the story, calls this front-loading the brief.
Only when they were clear did they go back to the AI. The result was a couch in a box, light enough to carry under one arm, and it ended up in a museum exhibition in Copenhagen.
Same AI tool, different starting point. Different end.
The gravitational pull towards the average that AI forces on businesses is what we need to work against. Everyone has access to the same models, and left to itself AI drifts towards the middle of everything it has seen.
Ethan Mollick of Wharton puts it bluntly: if you simply automate, you end up with the same AI-built company as everyone else.
The people who get an advantage put human thinking first.
Jeetu Patel, who runs product at Cisco, describes AI agents as teenagers: supremely intelligent, with no fear of consequence, and making poor decisions more often than you would like.
You don't hand a teenager the car keys and walk away. You don't lock them in their room either. You give them direction, set the boundaries and then let them run.
That is leadership work, and it can't be handed to the teenager.
In practice, there are some key places that need human creativity and supervision.
The brief
Before anyone opens a prompt, get clear on what the thing is for. IKEA's conversation about campfires changed the result entirely. Ask what your team knows about the customer, the purpose and the trade-offs that AI can't know.
The moments that need a person
This seems obvious, but Goldman's examples are useful. Angry customers want a person, while embarrassed ones would rather deal with AI. One healthcare company makes sure a person books the first appointment after a cancer diagnosis, even though AI could. Map your customer's journey and decide which moments belong to people.
Entry-level work
If AI takes the entry-level work, where will your future leaders learn? Salesforce answered by hiring a thousand graduates and interns this year, and Goldman says she has never learned so much from interns.
Human rough edges
Wade Foster, Zapier's CEO, turns down AI's polish to keep his rough edges. His rule is that the author should work harder than the reader.
He also tells people how much of himself is in a piece, from "a quick query, take it if it's useful" to "I stand by this 100 per cent", so they know how much weight to give it.
What not to build
When code is close to free, Patel says, choosing what not to build matters more than choosing what to build.
Cisco expects AI to write 70 per cent of its products by the end of next year, and it will have more engineers at the end of it. Patel wants every line an agent writes tied back to a person who owns it.
What you measure
Mollick warns that if you leave the KPIs unchanged, you get a hundred times more PowerPoint.
He asks two questions: what have you stopped doing because of AI, and what impossible thing are you building? That is a leadership decision, not an AI decision.
The key question is: what leadership work is needed before you fire up Claude?
So everyone has the same AI.
How you use it decides whether you get the boxy sofa or the couch in the museum. And whether you drown in sloppy AI PowerPoints or find a new business model.
Sources: Harvard Business Review's IdeaCast series on how AI is changing leadership: talent (Paula Goldman), communication (Wade Foster and Matt Abrahams), innovation (Jeetu Patel) and strategy (Ethan Mollick).