Good developers' caution about agentic development stems from knowing that built isn't the same as working in production. Despite being early adopters of new tools, many excellent developers hesitate to fully embrace coding agents. The obvious explanations—love of coding or job security—are only part of the story. A deeper fear lies beneath: without experience building systems this way, how much can we trust them? Experienced developers value judgment gained through years of hands-on work. They still get more useful answers from humans than models for difficult questions, as people can challenge premises, recognize patterns from past failures, and spot organizational problems hidden in technical ones. They are not hesitant to use coding assistants, but reluctant to let agents handle all execution without proper guardrails. Nearly every system in production today was built the old way, by hand, by trusted people. Everything built now is the new way, and we haven't lived with that yet.
Thoughtworkers care deeply about building the right thing and building it right. Practices like continuous integration, pairing, automated testing, and microservices emerged from this mindset. Early coding agents made it easy to produce code unsuitable for production, written by those unfamiliar with the gap between building something and making it run safely. While tools have improved, experienced developers still ask: Is this the right behavior? Will it work as expected? Can it evolve? Does it follow hard-learned principles? What happens when something fails?
A week at the whiteboard, Giles Alexander brought senior engineers together to work entirely with agents—no human-written code. The point was to explore exactly this challenge.
Source: Hacker News · Summarized by HeadlinesBriefing