Frontend's Lost Decade Is Repeating Itself in the Age of Agentic AI

Mauro Bieg argues that the disruption AI is bringing to programming mirrors what JavaScript frameworks did to frontend work over the past decade: a textbook case of deskilling. Frontend was once a specialized craft demanding fluency in semantic HTML, CSS quirks across browsers, accessibility, progressive enhancement, and performance tuning. Frameworks like React turned the browser into a generic compilation target, letting any generalist ship features by gluing components together without understanding what runs underneath. The payoff for businesses was interchangeable engineers and weaker worker leverage; the cost was slower, less accessible sites and a generation of developers who never learned the fundamentals.

Agentic coding fits the same pattern but with a sharper edge: it’s a non-deterministic abstraction. Compilers produce predictable output; LLMs produce different code from minor prompt or model changes, and the omitted details get filled in by training data and context guesses. The common comparison to junior engineers breaks down because juniors actually learn, while LLMs require endless tweaking of AGENTS.md and SKILL.md files to behave consistently. Bieg sees this as a continuation of the Stack Overflow era, where copy-paste programming let people produce code that sort of works without understanding it.

The piece captures a recognizable grief among practitioners whose hard-won expertise is being devalued by the market, while also acknowledging the legitimate framing of abstraction-as-automation. The honest question isn’t whether the abstraction is useful but which details it hides, who decides those details are unimportant, and what leaks through when they inevitably matter.