I recently joined Thomas Marbella on the RichDevPoorDev podcast to discuss a question that is becoming increasingly relevant for software developers: what happens when AI stops being just a coding assistant and starts operating the development workflow?

The conversation builds on my article, “AI in the Driver’s Seat: Build the Tools, Let the AI Operate Them.” A year ago, I mainly used AI to write individual functions faster. Today, my AI tools can work with issue trackers, repositories, tests, production logs and deployment pipelines. A task can move from a Jira ticket to a tested pull request without me typing the implementation myself.

That does not make the developer irrelevant. It changes where our value lies.

The important skills are increasingly the ability to define the right outcome, provide useful context, design reliable tools, establish guardrails and review the result critically. The bottleneck is no longer typing speed; it is clarity, architecture, judgement and responsibility.

Thomas and I also discussed the uncomfortable questions: Where are human approval gates still necessary? Who reviews the growing volume of generated code? What happens to junior developers? And does a fivefold productivity increase create more freedom—or simply five times the expectations?

A developer’s market value will depend less on how quickly they can write code themselves and more on how reliably they can build a system in which AI can work.

For developers who want to remain valuable in the market, “I can prompt” will not be a meaningful differentiator. The real proof is a reliable delivery system: shorter lead times, reproducible workflows, strong tests, safe integrations and consistently good results.

The episode is in German and available on YouTube: Agents vs Entwickler: wo sind HumanGates wichtig?

Related article: AI in the Driver’s Seat: Build the Tools, Let the AI Operate Them.