Is AI a threat to developers, or a career opportunity?

Reducing the AI debate to “will it take our jobs?” misses a practical change in development work. Writing code is one responsibility among several. Choosing the right problem, designing a useful flow and keeping a system maintainable still require decisions.

Acceptance criteria give speed meaning

An assistant may produce a convincing membership screen quickly. That screen does not decide what happens when a password-reset link is used twice, when a session expires or which records a user may access. Those behaviours need explicit requirements.

Start with a small task and describe its inputs, outputs and failure conditions. Instead of requesting “a form”, explain what an invalid submission should display and when a record counts as saved. That makes the response easier to evaluate.

Treat the first output as a review candidate

Running generated code is necessary, but understanding its behaviour matters too. Can you explain why it works? What does each dependency do? Does a duplicate request create a duplicate record? Does a service outage preserve the user's text?

Testing one behaviour in a small example is easier to review than asking for a complete rewrite. A limited change also gives you a clearer starting point when something breaks.

Use the time opened up by routine work

Reducing repetitive work can leave more attention for product decisions. Which information belongs first? Why do users hesitate at this step? Would simplifying an existing flow help more than adding another feature? The value of these decisions is independent of the number of lines produced.

I do not assume that automation improves every task. If using an assistant means abandoning review, an immediate speed gain can become maintenance work later. Understand a suggestion's assumptions before accepting it.

Keep learning active

Describe a problem in your own words before asking for a solution. Compare two approaches and ask when each one makes sense. Explain unfamiliar lines rather than memorising them. Repeating a response is different from adapting an idea to a new problem.

This article does not predict which profession will disappear. It invites a discussion about responsibilities we should deliberately own: validation, product choices, accessibility, maintenance and the user's experience.

Where do you use AI in your workflow, and which decisions do you prefer to make yourself? Bring a concrete example to the discussion.