The Post‑AI Era: How Large Language Models Will Transform Software Development

Matt Welsh, co‑founder of the AI application platform Fixie and former chief engineer at Google and Apple, observes that after more than fifty years of software development humanity is still bad at programming and may remain so for another fifty years.

He warns that the future of software will be dominated by large language models (LLMs) and asks critical questions: how will we work with LLMs, what will engineering teams look like, and is there an essential human element that must be preserved?

— Matt Welsh, AI application platform Fixie co‑founder

Welsh predicts the "end of programming" and stresses the need to understand what the software industry will become after AI. He highlights a "magic prompt" discovered in early 2023—"let's think step by step"—which makes LLMs switch to a computational mode rather than merely regurgitating answers.

He describes prompt engineering as an experimental discipline that requires trial‑and‑error to build reliable prompt manuals. The difficulty lies not in making LLMs fail, but in diagnosing why they deviate from human intent and determining the next steps.

In the envisioned post‑AI software teams, only passionate product managers and code reviewers remain human, while most other roles are automated. This shift could democratize computing power for a broader population.

Welsh cautions that software development is more than writing code; substantial work precedes coding, such as requirements analysis and design. He cites Frederick P. Brooks' distinction between essential (conceptual) and accidental (implementation) complexity, noting that LLMs will revive the need for rigorous requirement analysis, precise specifications, and software verification.

Ultimately, LLMs will not end programming but will revive fundamental software‑engineering techniques, offering powerful computational capabilities that should be accessible to everyone rather than a specialized elite.