Data · AI · Law · Ethics

The magazine for decision-makers

AI that writes code: is the programmer really becoming a supervisor?

Coding agents are changing the work of developers. Less time on the mechanical production of code, more responsibility for architecture, verification and quality.

by

·

·

1–2 minutes
Righe di codice colorate su un monitor, in primo piano

Coding assistants no longer just complete a line of code. They can propose changes across multiple files, generate tests, explain repositories and tackle increasingly complex tasks. This does not mean the developer is disappearing. It means their job is changing.

From autocomplete to agents

The first phase of AI for developers was mainly about suggestions: completing functions and boilerplate. The new phase is more agentic. The system receives a goal, explores the available context, modifies code and can iterate based on tests and errors.

The bottleneck shifts

If producing code becomes faster, framing the problem well becomes more important. Architecture, requirements, security, quality and domain understanding become even more decisive. The ability to type quickly matters less than the ability to understand what should be built.

A supervisor, but a skilled one

Calling the developer a supervisor can be misleading if we picture a passive role. Judging automatically generated code requires solid skills: recognizing vulnerabilities, poor abstractions, unnecessary dependencies and behaviors that work in testing but fail in real-world cases.

More code means more responsibility

The ease of generation can increase the amount of software that has to be maintained. The risk is mistaking local speed for overall productivity. Code that is generated quickly but is hard to understand can turn into technical debt just as quickly.

Where AI really helps

  • boilerplate and repetitive transformations;
  • first drafts of tests and documentation;
  • exploring unfamiliar codebases;
  • narrowly scoped refactoring;
  • prototyping and comparing alternatives.

A new metric for teams

Measuring success by counting lines of code or pull requests makes even less sense than before. Metrics should focus on delivery time, reliability, incidents, maintainability and value delivered. AI can increase output without necessarily increasing the outcome.

In brief

The programmer is not simply becoming the supervisor of a machine that writes code. They are increasingly becoming the designer of the problem, the context and the criteria by which the result is assessed. In a world where producing code costs less, understanding which code deserves to exist becomes even more important.

Leave a Reply

Your email address will not be published. Required fields are marked *