Your imaginary customers have excellent manners

A simulated customer can give you an answer whenever you want. A real customer can surprise you, ignore your product, or care deeply about the detail your team dismissed. This issue pairs the appeal of AI simulations with a reminder about where useful knowledge comes from. We like tools that help us ask better questions. We get twitchy when they make the people with answers seem optional.

A synthetic audience is easy to convene

Lenny Rachitsky points to teams using AI simulations to explore product ideas, changes to user flows, and copy.

Our take: It’s an appealing way to surface possible reactions before talking to people. The trap is letting a plausible answer become evidence of what customers want. If the audience exists because you described it to a model, some of the apparent insight may be your assumptions coming back in a nicer outfit. Use that output to improve the questions you take to actual people.

The expert who feeds the system still matters

In McKinsey’s edited interview transcript, Erik Brynjolfsson argues that firms should keep investing in experts whose knowledge helps AI improve other workers’ performance.

Our take: “Everyone can now perform like our best people” is a strange reason to stop investing in the best people. Someone has to handle the unfamiliar case, notice that the standard answer has gone stale, and create the next good example. We would reward that contribution explicitly. A company that spreads expertise while starving its source is borrowing against tomorrow’s competence.

A transformation plan can silence the people it needs

In a LinkedIn opinion post, Stuart Winter-Tear argues that fear about job security can discourage employees from sharing the knowledge an AI transformation needs.

Our take: If your process map depends on people telling you where the work really breaks, job anxiety is an information problem as well as a human one. Involve the people doing the work before presenting a finished redesign. Explain what has been decided, what remains open, and where their judgment can change the proposal. “We’re here to augment you” is a poor substitute for answers.

One useful move

Working on a new service, event, or internal resource? Give your usual AI tool the short description and ask:

I’m exploring {{topic_or_problem}} for {{intended_audience}}. Write five neutral questions I could ask them to understand how they handle this problem today. Ask about specific recent experiences. Don’t ask whether they like my idea, and don’t answer on their behalf.

Choose the two questions whose answers you most need. Put them in the feedback form or conversation notes you’re already preparing.

Something fun

Turn your chatbot into a very small text adventure:

Run a text adventure set in a supermarket just before closing. My goal is to buy one banana. Every aisle contains an unnecessary mystery. Give me three choices at each step and wait for my move. Keep the stakes low and the atmosphere excessively dramatic.

Somewhere near the frozen peas, a trolley knows too much.

Built in the past two weeks

We published two Mastermind guides: AI Fluency Levels and Build and Maintain Team and Organization AI Toolkits. One gives people a vocabulary for developing how they work with AI; the other explains how to turn useful prompts, context, and procedures into shared assets a team can maintain.