Idea sessions that go somewhere

Brainstorming with AI has a reputation problem: ask for ideas, get twenty bullet points of beige. The tool isn't the issue — the session design is. A good idea session has a shape: wide, then weird, then narrow. This lesson gives you that shape as a repeatable practice, plus the moves that turn "a list of ideas" into "the idea, chosen."

Go wide with quotas, not vibes. "Give me ideas" produces the average. "Give me twenty ideas: five obvious ones to get them out of the way, five for a tight budget, five that need a partner, five that would only work for someone exactly like me: [context]" produces a landscape. Quotas per category force the model out of its comfortable center — the same options-upgrade from Foundations, industrialized.

Then go weird on purpose. The obvious list is warm-up. Follow with deliberate distortion: "now five ideas that would be terrible advice for most people but might work for me because [constraint]. Five that steal a mechanic from a completely different industry. Five my most skeptical friend would propose." Weird rounds don't usually win — they dislodge the assumption the winning idea was hiding behind.

Narrow with criteria you name out loud. Before reading the long list twice, write the filter: "score each against: under ten hours to try, uses what I already have, failure costs nothing but the time. Top three, with the single biggest risk of each." Naming the criteria is the part most people leave out — and it is where the session stops being entertainment and starts being a decision-in-progress.

Interrogate the survivors. The top idea earns a hostile round before it earns a plan: "argue this fails: name the three most likely failure reasons, who tried something like it and struggled, what I'm underestimating." It's the same role-switch you used in the practice-partner job — the model plays your toughest critic instead of your cheerleader, and plays it without mercy or memory of having praised the idea a message ago. Survivors of the hostile round go to the plan-skeleton job from earlier in this course; casualties saved you weeks.

Keep an idea garden, not an idea graveyard. End every session by asking: "which three discarded ideas are worth revisiting in six months, and what would need to change?" Park those in a note with their conditions. Ideas rarely die of being wrong; they die of wrong timing — the note is how good-but-early ones survive your own filter.

Group sessions work too — with one change. Running this shape before a team meeting turns the meeting from generation to selection: bring the twenty quota'd ideas printed, spend the hour on criteria and the hostile round together. People argue better against a list than at a blank whiteboard — the awkward silence where everyone waits for someone brave has already been handled by the machine, and the human hour goes to the part humans are actually better at: knowing what this particular team can pull off.

One mistake to skip: brainstorming without a decision date. A session that doesn't end in "one idea gets ten hours by Friday" produces the pleasant illusion of progress — the feed-scrolling of productivity. Wide, weird, narrow, hostile, then committed: the last step is the one that happens off-screen, and it's yours.