Vertical slices beat phases

Phased delivery feels orderly on a plan and produces, for most of its duration, nothing a business stakeholder can evaluate. That is its actual failure mode.

The problem with layers

Building the data model, then the services, then the interface means the first honest feedback arrives after the most expensive decisions are already committed. Every assumption made in month one is validated in month six, when changing it costs the most.

A slice that goes all the way through is worth more than three layers that go most of the way across.

What a slice looks like

One real workflow, from the screen a person touches through the domain logic to the row that is written and the report that reads it. Narrow enough to finish in a week, complete enough that someone can do their actual job with it and tell you what is wrong.

• It runs in a real environment, not a demo branch.
• It uses real data shapes, even if the dataset is small.
• It includes the unglamorous parts: permissions, errors, empty states.
• Somebody outside the build team can complete a task with it unaided.

What it changes

Estimation gets better because you are extrapolating from delivered work rather than from a plan. Scope conversations get concrete because there is something to point at. And the architecture gets tested against reality early, while changing it is still cheap.