11What changes for development teams

When an AI writes the code, the work of a development team moves to deciding and verifying. This chapter points out what changed on this project. What it means for teams and their leaders deserves an article of its own.

In twelve weeks, the product grew to about 24,000 lines of Java, more lines of tests than that, and the HTML, JavaScript, CSS and SQL around them. I wrote no product code. My time went to what the product must be, which architecture it needs, which method builds it, how it should be tested and how it stays compliant. Writing code was never what limited the project.

What changed?

One contributor across every layer

The simplest page in the product, the one that shows a patient's record, takes seven skills to build end to end. Teams have long split such work between specialists, with a hand-over at each boundary. Here, Claude covered every layer, with a fluency across all of them that few developers combine, and there was nobody to hand over to. Scrum has long asked for a cross-functional team without asking each person to master every skill (Scrum in Name Only); an AI fluent in every layer brings that breadth with the tool.

The shape of a team, before and with an AI fluent in every layerThe seven skills one simple page needs. Until now they were split between front-end, back-end, database and security specialists, with a hand-over at each boundary. With an AI fluent in every layer, Claude writes across all seven, and people decide and verify: architects, domain experts, a product owner, and engineers who can review every layer.UNTIL NOW · SPLIT BY LAYERONE SIMPLE PAGEWITH AICSSJavaScriptHTML templatesServer-side JavaApplication serverDatabase and SQLIdentity protocolFront endBack endhand-overDatabasehand-overSecurityhand-overClaude writesPeople decide and verifyArchitectsDomain expertsA product ownerEngineers who reviewevery layer
Figure 12. The seven skills behind one simple page: split between specialists until now, with a hand-over at each boundary; with an AI fluent in every layer, people decide and verify.

Reading and judging became the core of the work

Someone had to notice that the persistence framework was the wrong choice, settle nearly 900 review findings, trace a bug that read the wrong column and passed every test, and check that the package deployed on the server was the one just built. That takes people who could have written the code. For a medical device, the standards ask for exactly that: verification by people competent to do it.

Knowledge of the domain and of the product counts for more

What the product must do, and why, came from the people who know the work and from a product owner. The team this project points to has architects, domain experts and a product owner, and people able to read every layer of the code critically.

What comes next?

These changes raise questions a chapter cannot do justice to: how teams are composed and grown, where the next seniors will come from once juniors write less code, what the change asks of leaders, and why it worries people, who deserve a straight answer. I have been through three shifts of this kind before, and they will be the subject of a separate article on this site.

For your own project

  • Map the skills one feature needs end to end, and count the hand-overs.
  • Keep, and develop, the people who can read and judge code across the whole stack. They become the constraint.
  • Start the conversation about roles early, with the people concerned.