SOLUTIONS
“Does anyone remember why it works like this?” Your code should.
The colleague who knew is on holiday. The wiki page is from 2021. And the test is called test_case_7. FACET writes down what a feature is supposed to do — right next to the code that does it and the test that checks it.
01 · BUILD CONFIDENCE
Which edge cases do our tests actually cover?
- Testing and QA“All green.” And the customer still calls.Our tests are green. But do they test what would actually hurt?
- Bug investigation and review“That line can go.” Are you sure?Can this line go — or does something depend on it that nobody remembers?
- AI-built productsThe agent built it. Who checked it?The demo works. But what happens when real customers start clicking?
02 · SHARE UNDERSTANDING
Why do I have to explain this every single time?
- Documentation and supportThe help article says 14 days. The code says 30.Why does the help article still show the old deadline?
- Onboarding and handoffs“Best ask Thomas.” Thomas left in March.How long until the new hire can work without asking around?
- Product management“It’s done.” Does it do what we wanted?Is it done — or just done coding?
03 · NAVIGATE COMPLEXITY
What else breaks if I change this?
- Monoliths and legacy systemsNobody wants to touch the pricing logic. And everyone has to, eventually.Who still dares to touch the pricing logic?
- Enterprise and distributed teamsYour release was green. Ours is broken now.We change something — and find out in production that another team depended on it.
IS FACET A FIT?
A weekend project doesn’t need this. Your product in three years does.
When the extra structure can pay off
The case is strongest for long-lived products: hard-to-navigate monoliths, multiple teams and domains, repeated handoffs or frequent agent edits. What matters is whether knowledge will be reused and someone will maintain it.
When a README and tests may be enough
For small projects a person or agent can understand in full, and for throwaway prototypes, FACET may create more maintenance than value. If existing tools already connect rules, code and evidence reliably, another layer may be unnecessary.
THE FIRST STEP
Take the question that bugs you right now. One hour is enough for an answer.
No rollout, no big project: one feature, one question, one hour. Afterwards you know whether the answer comes faster next time — or whether you don’t need FACET at all.