As coding AI develops rapidly, an argument has emerged that a more important capability in software development is understanding the industry and the work itself rather than programming.

On June 1 local time, online outlet Gigazine reported that software developer Aaron Brethorst (아론 브레트호르스트) pointed out that as AI writes code in place of people, the value of expertise to judge whether a system meets real business requirements increases.

Brethorst said the difficulty in software development lies more in accurately understanding the target industry and work structure than in writing code. Even for a payroll system, he said, understanding tax rates, deduction conditions and adjustment rules by pay period matters more than creating calculation code to determine whether the system operates properly.

He gave an example of a dispatcher who worked in the logistics industry for 15 years and a strong software engineer using the same AI coding tool. The dispatcher can judge whether an AI-made logistics system fits on-the-ground needs even without programming skills. The engineer, he said, may be able to assess code quality but may not be able to tell whether the system meets real business requirements.

Brethorst said code is closer to an output that documents industry knowledge. He also said that as agentic AI makes it possible to build software without directly constructing an operating model, the link in traditional development methods, where people had to learn domain expertise and code together, has weakened.

The analysis is that this change is also shifting the relative value of engineers and field experts. Until now, engineers have built systems by collaborating with experts and going through trial and error in real operating environments. Field experts, by contrast, rarely built systems themselves because it took a long time to learn how to create reliable software. But as AI has sharply lowered the cost of turning ideas into working software, engineers' technical edge has weakened in relative terms, while the value of specific on-the-ground knowledge has risen, he said.

An Anthropic hackathon was presented as an example of this claim. The event, which pitted participants against each other in using the latest AI models, drew 500 people and most were developers, but 3 of the 5 winners had no experience launching software. Systems researcher Dexter Hardley (덱스터 하들리) said the result was a case where domain expertise outpaced coding ability.

Brethorst said areas where experienced engineers should invest time include deep understanding of real industries and work processes, specialized equipment and regulatory systems. The value of implementing in clean code has declined, while knowledge that deeply understands real work and is validated in practice remains scarce, he said.

A counterargument also emerged that field experts will not immediately succeed in software development. On Hacker News, an opinion was raised that the ability to validate whether a system output is correct and the ability to instruct AI to produce correct output in the first place are different. It also said that even with deep knowledge, experts in a specific field may struggle to clearly organize rules learned through long experience into tests and requirements that AI can understand.