InfoQ talk discusses the AI coding supervision paradox

On October 30, 2026, InfoQ published a talk by Alasdair Allan in its AI/ML section, examining how AI coding tools are being adopted and what that means for engineering teams over the long term. Drawing on empirical research from METR and Anthropic, Allan articulated what he called the "supervision paradox": dependence on AI tools disrupts the skill development of junior engineers, risks the loss of institutional context, and threatens the long-term engineering pipeline that will be needed to oversee autonomous systems effectively. According to Allan, the mechanism is straightforward in principle. As teams lean more heavily on AI coding assistants for day-to-day work, entry-level engineers are shielded from the hands-on problem solving that builds deep technical judgement, so the bench of experienced engineers that organizations will rely on to supervise increasingly autonomous systems is not being replenished. Allan pointed to the METR and Anthropic findings as evidence that this pattern is already visible in practice rather than merely theoretical. The talk frames the issue as a trade off between short-term productivity gains from AI tooling and a longer-term capability gap in supervision, without prescribing a specific remedy. That, Allan suggests, is the next question the field needs to take up: how organizations can capture the benefits of AI coding tools while still growing the engineering talent they will need to govern autonomous systems in the years ahead.