Artificial intelligence is increasingly able to run code and even make revisions based on the results.

Gigazine reported on March 20 that web developer Simon Willison (사이먼 윌리슨) named this “agent engineering” and said a new paradigm has arrived in which AI leads the coding process.

Coding AI is transforming software development by moving beyond simple code writing to execution and verification. Willison said AI developers should now focus on deciding what to build, preparing tools and validating results.

Willison stressed that lower costs for writing code are at the heart of agent engineering. In the past, writing hundreds of lines of code took more than a day, but repetitive work has been sharply reduced thanks to coding AI. However, the cost of producing good code still remains. He explained that code accuracy, ease of maintenance, testing and documentation are still the responsibility of human developers.

In agent engineering, the principle of “accumulate what I can do” is important. It is not just a theory of what you can do, but actually experiencing executable code that becomes a powerful weapon, he said.

Willison also countered concerns that AI lowers code quality, saying that is only the result of using AI incorrectly. He said AI should be used to reduce technical debt and try diverse approaches. For example, when judging whether a specific technology is suitable, using AI to generate many prototypes can reduce errors at the design stage.

But misuse of AI also needs to be guarded against. Willison called sharing AI-generated code without verification an “anti-pattern.” He said developers must personally check AI-written code and be able to review it in small units. He stressed that while AI is innovating coding, ultimate responsibility still lies with developers.