AI made answers cheap.Judgement just got expensive.

Ask an AI tool for a commercial laundry’s value proposition and it’ll hand you one in seconds. “Reliable linen solutions that let hospitality teams focus on their guests.” Respectable. Grammatical. Usable, word for word, by every competitor in the country.

Spend twenty more minutes on the prompt and you’ll get a punchier sentence. You still won’t know which part of the service a hotel manager would actually switch suppliers for.

That’s the trap. It’s never been easier to produce something that looks finished while the important question is still open.

Here’s my position. AI should make your team faster at the work that doesn’t need judgement, and deliberately slower at the decisions that do. Most teams are doing the opposite.

Making options got cheap. Choosing didn’t.

The best evidence so far comes from inside a real company. In a field experiment with 791 professionals at P&G, published in Organization Science in June 2026, people worked on real product innovation challenges alone or in pairs, with or without AI. Individuals using AI produced proposals that matched the quality of teams working without it.

That’s the fast half. Then the slow half. People using AI were worse at picking their own best idea. Roughly 37% of them chose their strongest proposal, against about 50% for pairs working without AI. Their picks still came out better on average, because AI lifted the quality of the whole pool. But the choosing got worse.

Read that again, because it’s the whole argument. Generating options and choosing between them are two separate skills. The machine helped with one. On the evidence of this study, it made the other harder.

Earlier research from Harvard Business School and BCG with 758 consultants points the same way. AI lifted performance on tasks inside its capabilities. On a task outside them, consultants using it were 19 percentage points less likely to get the right answer, and the two kinds of task looked similarly hard. Summarising interviews, checking a brief for contradictions, generating creative routes and making a factual claim about a customer are four different jobs. Confidence in the first shouldn’t travel to the fourth.

So when output gets cheap, the scarce thing is the person who can say which output is right, and why. That’s a strategist. It was always the expensive part of the job. Now it’s most of the job.

Give it a decision, not a blank page

Back to the laundry, an illustrative one. The owner needs to decide what to lead with when pitching independent hotels: lower cost per room, better linen, or confidence about what’s arriving before check-in.

Give the tool a small, honest evidence pack. Two housekeepers have described finding out too late that their linen order would be short. Dispatch now sends a count of available items the afternoon before delivery. The business can’t guarantee every item, has no verified competitor pricing, and no customer has mentioned how the sheets feel.

Now the brief writes itself. Develop three ways to explain the service to a hotel manager, each with the customer problem, the supporting fact and the claim we’d have to verify, plus one reason the manager might stay with their current supplier.

The model will offer “Never run short of linen again.” Clear, specific and wrong. An advance count gives the hotel time to respond to a shortage. It doesn’t abolish shortages. Publish that line and an operational improvement becomes a promise the team can’t keep.

“Know what’s coming before tomorrow’s check-ins” fits the evidence better, and still needs checking with operations and a few buyers. And if hotels are tendering on price and the business has a verifiable cost edge, the best answer is the least exciting sentence on the page.

Judgement includes binning the line you liked.

What it looks like inside a team

When Hum built an account management session for Rare, AI was part of the brief from the start. The premise was simple: AI can make output cheaper, but it can’t decide what matters. As production gets faster, more of an account manager’s value shifts to knowing what should be done, why, and how to hold the commercial conversation around it.

The team left with a Suit’s AI Playbook built for that job: strip out low-value effort, protect the thinking. Managing Director Simon Stewart called the results “clear and immediate”.

That’s the stand in practice. Use the time AI gives back to ask sharper questions, not to produce a prettier document for a meeting nobody needed.

What to do on Monday

1. List the work your team does each week. Mark each task as production (drafting, summarising, resizing, formatting) or judgement (choosing, prioritising, claiming, promising).

2. Push AI hard into the first column. Track the hours saved.

3. For anything in the second column, write down what would make a plausible answer wrong before you ask the tool anything.

4. Feed it verified facts only, with assumptions labelled. A generated persona isn’t a customer interview.

5. Before choosing between AI-generated options, put the best two in front of real buyers or the people who’d have to deliver the promise.

The machine will give you ten good answers. Your job is knowing which one is true.

Sources & notes

• Dell’Acqua et al., “The Cybernetic Teammate”, Organization Science, published 12 June 2026 (field experiment with 791 P&G professionals; about 37% of participants using AI picked their best idea against about 50% of pairs without AI). Innovation tasks at one company, not campaign sales effects.

• Harvard Business School AI Institute, “Navigating the Jagged Technological Frontier” (758 BCG consultants; on a task outside AI’s capabilities, AI users were 19 percentage points less likely to get the right answer)

• Rare case study, Hum, including Simon Stewart’s quote

• The commercial laundry is illustrative, not a Hum client or a reported AI experiment.

Continue with The best thing about your business never leaves the building. If you want a next step, see Hum’s fractional CMO page.