As next-generation artificial intelligence (AI) models emerge and coding agents begin producing applications that actually work beyond merely plausible code, the development field is being seen as reaching a turning point.

Gigazine reported on April 6 that software engineer Simon Willison (Simon Willison) said in a podcast, "Previously you had to review the output code very carefully, but recent models are starting to produce results that work as instructed."

Willison stressed that the reliability of outputs has changed significantly. In the past, many generated codes were difficult even to run, but cases of properly working apps are increasing, he said. He explained that this is making it a reality for AI agents to take on not only writing code but also some of the execution and testing.

Development productivity has also increased sharply. He said he has been able to write up to 10,000 lines of code a day with AI help. He cited as an advantage that code quality can be verified relatively clearly by whether it runs. He said it is much harder to evaluate AI results in areas where it is difficult to judge correct answers, such as essays or legal documents. He also mentioned that there are quite a few cases in which lawyers fail to filter out AI falsehoods.

As development speeds up, bottlenecks are also shifting. In the past, it took weeks from delivering specifications to implementation, but now results sometimes come back within hours. Instead, the next task is moving to testing, and verifying rapidly generated outputs is emerging as a new burden.

The way prototypes are made is also changing. Willison said that on the premise that the first idea in product development is always wrong, he tests by making multiple prototypes with different ways of working when designing a single function. He assessed that tools such as ChatGPT and Claude quickly present persuasive results, especially in user interface (UI) design. He added that selecting the best option still remains a human task and that traditional usability testing is still important.

But as efficiency has improved, the burden has also grown. He explained that fatigue can quickly accumulate when running several coding agents at the same time, stressing, "Using AI tools well is harder than you might think and requires a lot of practice." He said mid-career engineers in particular could face the biggest challenge amid role changes driven by automation.

Willison cited "agency" as what separates humans from AI. He stressed that it is important to build organisational skills and problem-solving ability and to actively use new technologies, adding that in the AI era, the ability to handle tools itself will be a competitive edge.