What changes when nobody on the build is learning on your time

The questions come first

The visible difference in week one is the questions. Where does this data come from, who is allowed to change it, what happens when it is wrong, and who does this by hand today. Those are not curiosity. They are the questions that decide whether the thing still works in month nine.

An engineer who has not shipped an operational system before has no particular reason to ask them, because nothing in their experience has yet made them expensive.

Refusal is part of the work

The most valuable thing a team can say early is that a requested feature will cost more than it returns.

A request that arrives with an assumed solution attached is the common case. Unpicking the request from the solution, and sometimes declining the solution, is where most of the saving in a project is found. It mostly requires having been wrong before in the same way.

Scope shrinks, not grows

Experience tends to make a proposal smaller. The parts that get removed are the ones that looked cheap and are not: the configurable rules engine, the second integration nobody has asked for, the admin screen for a table with nine rows in it.

Fewer people, fewer seams

A larger team is not simply more capacity. Every additional person adds an interface, and the defects that survive to production are mostly born at those interfaces rather than inside any one part.

A small team that holds the whole system in mind can make a decision on one side of it knowing what it does to the other, and that is not a process which can be written down and handed over.

The second version is planned

Nothing survives contact unchanged, and a build that assumes it will is the one that becomes impossible to alter. Planning for the second version means keeping the pieces separable and the decisions written down somewhere the next person will find them.

Estimates include the dull parts

The estimate that comes in low is usually the one that priced the interesting work. What runs over is the rest of it:

• migrating the data that already exists
• the states nobody described
• access control and audit
• handover to the people who run it

None of that is optional, and all of it is invisible in a demo.

What it costs

A senior-only team is more expensive per week and it is not always quicker to start. The trade is fewer weeks, and far fewer of the weeks that are spent undoing something.

What it does not buy

It does not buy certainty, and it does not remove the need for the client to make decisions. What it removes is the category of mistake that is only obvious to somebody who has already made it once.