Plan in vertical slices
A layered plan hides all the integration risk until the end, and the estimate is a single all-or-nothing bet. Slicing vertically — one thin path through every layer — produces shippable increments, earlier feedback, and estimates that can be checked slice by slice.
Procedure
Take the smallest user-visible outcome: "a logged-in user sees their own name on the dashboard." That is slice one.
Implement every layer needed for that one outcome — a real row in the DB, one endpoint, one rendered component. Resist stubbing layers you "will need anyway."
Ship the slice (behind a flag if needed) and observe it. Deploy it, do not just merge it.
Estimate the next slice now, using the first as a reference class. Slice N is usually cheaper than slice 1 because the plumbing exists.
Add slices one at a time, each end-to-end. A slice that only touches one layer is horizontal and does not count.
Keep a walking-skeleton slice zero: the thinnest possible path (health check through deploy through UI) to prove the pipeline works before features pile on.
Track slices as checkboxes in
notes/slices.md, so partial delivery is visible:- [x] slice 1: show own name - [ ] slice 2: edit own name - [ ] slice 3: avatar upload
Pitfalls
- Calling a layer-complete milestone a slice; "all the models are done" ships nothing a user can touch.
- Slices so thin they are chores (a whole slice for a single field) — merge them.
- Deferring the walking skeleton, so integration surprises arrive after most of the estimate is spent.
- Building a layer "properly for the future" inside slice one, which re-creates the big up-front risk.
- Estimating all slices up front to the same precision; later slices should be estimated more loosely and refined as you go.
- A slice that cannot be demoed on its own; if you cannot show it, it is not vertical yet.
Verification
grep -c '\[x\] slice' notes/plan.md
# passes when at least one slice is checked and every slice spans the full stack
Report the slice list, which slices are shipped and observable, and the re-estimate of the next slice from slice one's actual cost.