Guides · Updated 2026-11-03
What comes after spec-driven development
Spec-driven development stops at the merge. What comes after it is outcome-driven development: every change ships as a bet with a target, behind a switch, and gets a verdict from real usage — proven, disproven or unclear.
Spec-driven tools give a coding agent a persistent plan: a spec, then tasks, then code, with a person reviewing between steps. That fixed the first problem of agentic building — agents drifting from intent. It did not fix the second: nobody can say whether the feature they shipped mattered.
The missing half
Building is cheap now. Knowing what was worth building is not. A spec says what to build; it says nothing about how you will know it worked, what it cost, or what to do if it did not. Those questions live in analytics, flags and experiment tools that never see the spec.
Outcome-driven development in practice
Each piece of work is shaped as a bet: the number it should move, for whom, and by when. It ships behind a flag you can switch off without a deploy. Usage is measured against the bet, and the bet gets a verdict. The verdict feeds the next decision.
With Golden Frijoles, your own coding agent runs that loop. In Claude Code:
claude plugin marketplace add golden-frijoles/skills
claude plugin install golden-frijoles@golden-frijolesQuestions
Is outcome-driven development a replacement for spec-driven development?
No. It starts where a spec ends: the spec says what to build, the outcome loop says whether building it paid off.
What does an outcome loop need that a spec tool lacks?
A target for each change, a switch to turn it off, measurement of real usage, and a recorded verdict.