Tasks that are still expensive to generate

Why the rung fell out, stated as economics

The tasks traditionally handed to a new engineer were selected for one property: low blast radius. A CRUD endpoint. A test for an existing function. A copy change. A small bug with a clear reproduction. The work was real but cheap to get wrong, so a beginner could do it and a senior could check it in a few minutes.

Those are precisely the tasks that now take a model thirty seconds. So the rational allocation, task by task, is to not assign them, and the sum of those individually rational decisions removes the path by which anyone becomes senior.

This is not an attitude problem to be fixed by exhortation. It is a structural incentive, and the only thing that works against it is designing different tasks.

The property to select for

A good apprenticeship task in 2026 has this shape: the expensive part is figuring out what to do, not doing it.

Generation collapses the writing. It does not collapse diagnosis, negotiation with a stakeholder, tracing a behaviour through four services, or deciding which of three defensible options fits a constraint that is not written down anywhere.

Six that work

Assign the diagnosis, not the fix. “This endpoint got slow last month: find out why” is a week of real learning and cannot be typed away. The fix at the end might be four lines. The four lines were never the point.

Give ownership of something small and real, including the pager. A service with genuine users, a dashboard, an alert that rings. Nothing teaches operational reality like being the person who is woken. Pair the on-call, do not hand it over alone.

Put them on review first, presenting rather than receiving. Have the junior walk a senior through a change and field questions. This inverts the usual dynamic and is by far the fastest transfer mechanism, because the questions expose the gap immediately and in front of somebody who can close it.

Hand over an incident write-up. They interview the people involved, build the timeline, and write the document. They will learn more about how the system actually fits together in three days than in three months of feature work.

Assign the migration. Moving a table, retiring a field, changing a contract that three other teams depend on. This is mostly coordination and sequencing, the plan is the hard part, and it is useful work nobody wants.

Deliberately give them the ambiguous request. Not a ticket: a sentence from a stakeholder. “Finance says the export is wrong.” Their job is to come back with the questions, not the code. This is the specification skill from module four, learned the only way it is learned.

Make it a line in the plan

A team that intends to develop people and does not put hours in the plan will not develop people. The mechanism is not mysterious: under pressure, unbudgeted work is the work that gets cut, and development is always unbudgeted.

So: a named percentage of a senior’s time, in the sprint, with their name on it. Review presentations on the calendar. The junior’s diagnosis task sized as a real deliverable with a real deadline, not as slack. If it is not in the plan it is a stated value, and stated values lose to deadlines every time.

Cohorts beat individuals

Three juniors hired together learn faster than three hired separately across a year, and it is not close. They ask each other the questions they are embarrassed to ask a senior, which is most of the questions. They calibrate: “do you also not understand the deploy pipeline?” is an enormously useful sentence and it cannot be said upward.

If you are hiring one, see whether another team is hiring one, and start them the same week.

The honest part, which is about incentives

All of this costs senior time now for a benefit in eighteen to thirty-six months, and the person may leave before it lands. Every individual firm would rather hire somebody another firm trained. That is a textbook free-rider problem, and it does not resolve because people read an article about it.

Which means two things. If you are a manager, you are choosing to spend real money on a benefit that partly accrues to your competitors, and you should say that out loud when you propose it rather than pretending the return is captured. And if you are early in your career, some of this you will have to build for yourself, because a meaningful number of employers will not build it for you: take the diagnosis work nobody wants, volunteer for the incident write-up, ask to present at review.

The tasks that made people senior were never valuable in themselves. They were cheap containers for judgment. When the container got free, the judgment did not, but you have to put it in a new container on purpose.