Software's centaur age may last decades

Software engineer Sean Goedecke argues that engineers working with AI coding tools will outperform either alone for a decade or more, and that keeping agents aligned with organisational goals is their lasting job.

Sean Goedecke, a software engineer who writes essays about engineering work, asks how long human engineers and AI coding tools will work best as a pair. He compares the question with chess, where human and computer teams led for about twenty years. He concludes that this period in software may last a decade or two, though he calls the outcome uncertain.

• Goedecke traces coding tools from autocomplete in 2022 through chat assistants to AI agents that can run without supervision.
• He writes that most current agent errors are misalignment with organisational goals, such as over-engineering one area and under-engineering another, rather than ordinary bugs.
• He names alignment with organisational goals as the leading candidate for lasting human value, with taste as another.
• He advises engineers to keep human review in the loop and not to refuse AI tools.

An engineering team can check agent work against written organisational goals as well as against tests, because Goedecke places most agent errors there. Turning his career advice into a team review practice is an inference and is not the author's claim.

11 October 2026: AI agents depend on written context that goes stale, and the teams in these pieces review agent changes more firmly than they keep that context current.