We Compared 12 Mutations of One Product Idea
Workshop notes from running a meeting-notes product through four Idea Mutation Machine settings and comparing the 12 resulting directions.
More ideas are not automatically more choice. After several variations, people often stop comparing substance and start reacting to names. We wanted to see whether Idea Mutation Machine could help with the harder job: changing the shape of an idea without losing the problem underneath it.
Our starting point was intentionally ordinary: a tool that turns meeting notes into tasks. The page returns three cards for each direction-and-strength pair. We reviewed Tool / Small, Tool / Medium, Tool / Wild, and Weird / Small: twelve visible mutations in total.
The four directions we kept for a second pass
Input-to-output helper kept the original narrowness. Quality inspector introduced review criteria before task creation. Workflow mapper made blockers and owners explicit. Autopsy tool inverted the sequence by asking why the workflow might fail before proposing a build.
These were not finished product concepts. They were the four cards whose build notes would cause meaningfully different interfaces. The other cards were still readable, but they either widened the product too quickly or moved it into a different genre.
| Mutation | What changed | Why it survived |
|---|---|---|
| Input-to-output helper | Scope | Keeps one input and structured output |
| Quality inspector | Decision rule | Adds criteria and severity |
| Workflow mapper | Output | Adds stages, blockers, and owners |
| Autopsy tool | Sequence | Starts with failure modes |
What we discarded
Personal calculator imposed a score without identifying a meaningful quantity. Negotiation simulator added personas and objections that pulled attention away from the meeting record. The three Weird / Small cards were entertaining reframes, but receipt, ritual, and object-point-of-view interfaces did not improve the underlying accountability job.
Writing down these reasons was useful. The machine always returns polished cards; it does not mark a direction as unsuitable for the seed. A discard pile turns generation into an editorial process.
- Discarded: a score without a defensible measure.
- Discarded: a social mechanic before the private workflow worked.
- Discarded: a playful format that obscured owners and dates.
- Adapt later: constraint engine, once the core output is stable.
A better mutation prompt
The page does not accept a custom mutation instruction; it offers four directions and three strengths. We treated those controls as lenses, then compared the card fields: hook, build, audience, and validation signal. Keeping the original seed visible on every copied card made comparison easier.
For a second pass, we would take one surviving card into a separate brief and make it smaller. The machine is good at proposing directions, not reconciling them into a build plan.
How we would run the workshop again
Twelve mutations were enough. We would also invite one person to defend the original idea, because mutation sessions can reward novelty even when the simple version is better.
The machine is most useful in the middle of a process: after a problem is understood, before a solution is treated as inevitable. Used too early, it multiplies vague ideas. Used too late, it becomes decoration around a plan nobody wants to revisit.