Jesse Vincent (obra) / brainstorming
How do you stop an agent building the first idea you mention?
By keeping the conversation in design until the idea is worth building.
What to remember
- Eagerness is the failure mode it addresses. Code too early is expensive to unwind.
- Give constraints, not your preferred answer, or you will get your answer back.
- Real alternatives differ in tradeoffs. Three variations of one idea are not options.
- It leads naturally into writing-plans, which is where the decision gets recorded.
01What it changes in practice
The most expensive agent mistake is not a bug. It is building the wrong thing quickly and convincingly, because speed makes the wrong direction feel like progress.
Staying in design costs a few minutes and regularly changes what gets built. The test of a good session is that the plan at the end is not the idea you walked in with.
02Where it disappoints
It can generate options for their own sake. Two genuinely different approaches beat five near identical ones, so say how many you want and on what axis they should differ.
It also cannot decide for you. At some point somebody has to choose and own it.
Install and first run
- Install superpowers from obra/superpowers, published under MIT.
- State the problem and the constraints, and explicitly withhold your preferred solution.
- Ask for alternatives that a reasonable person would disagree about, not three versions of the same idea.
Questions people actually ask
- Should I share my preferred solution?
- Not at the start. Give the problem and the constraints, or the session becomes agreement rather than exploration.
- How many alternatives are useful?
- Two or three that differ in tradeoffs. More than that tends to produce variations rather than options.
- What comes next?
- Writing a plan, so the chosen direction is recorded before implementation.
Sources
Machine readable versions of this page