Why are good developers so wary of agentic development?

Experienced developers’ caution about agentic development is not simply attachment to writing code or fear of job loss. It reflects hard-won awareness that producing software is different from building something reliable, maintainable, and safe in production. Agents can generate code quickly, but teams still need confidence in its behavior, architecture, failure handling, and ability to evolve. The deeper concern is that few production systems have been built this way, so the long-term risks and effective safeguards remain uncertain. Skilled developers’ judgment - grounded in past failures and organizational experience - is valuable precisely because the route is not yet well mapped.

A week-long experiment in which agents wrote all the code showed humans spending about two-thirds of their time at a whiteboard defining the domain, boundaries, and plans, and one-third building a “harness” of tests, contracts, architecture checks, and delivery controls. The application ran, but builds were frequently red, undermining trust in both failures and successful checks. The team did not resolve whether parallel agents, system modularity, oversized changes, or slow checks caused the problem; DORA research likewise links AI adoption to faster delivery but lower stability. The argument is to involve experienced developers in experimentation, not dismiss them as laggards. Leaders must provide time and permission to learn, while developers must question familiar practices too: agentic development may make some safeguards more important and require others to change.