AI method · Claude Code · Personal · 2026
Stress testing a roadmap decision
Working alone means nobody argues back. I built a set of Claude Code subagents to do it, and learned what they are good for and what they are not.
- Role
- Built it, use it
- Stack
- Claude Code subagents
- Used on
- JourneyPal
- Roles
- PM · Critic · Devil's Advocate · Synthesizer
The problem with deciding alone
On a solo project every decision gets made by the person who already wants it to be true. There is nobody in the room to ask whether it should be built at all.
So I set up four Claude Code subagents with fixed roles. A Product Manager makes the case, a Critic goes after the reasoning, a Devil's Advocate argues the opposite, and a Synthesizer reconciles the three. Decisions go through them before I commit to building anything.
The set up
- Subagent
- A separate instance of the assistant given one job and one point of view, instead of one assistant asked to hold several at once.
- Orchestration
- Running several of those together and controlling what each one sees, so their answers can be compared rather than averaged.
Where it was worth it
Deciding whether JourneyPal was worth shipping at all, and working out which other approaches might support a business model that could actually hold up. Those are questions where being argued with is the whole value.
Where it was not
Design direction. It was not worth running layout and visual decisions through the set up, and I stopped doing it.
Useful for deciding whether to build something. Not useful for deciding what it should look like.