There is still no 'right answer' for developing software using AI

There is no established methodology for software development using AI, and everyone is proceeding by trial and error, but the technology is evolving so quickly that traditional expertise is no longer applicable. AI developer Scott Warner has published an article on his blog based on his own experience titled 'There is still no right answer for development using AI.'

Nobody Knows How To Build With AI Yet - by Scott Werner
https://worksonmymachine.substack.com/p/nobody-knows-how-to-build-with-ai

Warner used his AI assistant Claude to develop Protocollie , an application that explores and tests MCP servers, in just four days. Warner only gave instructions to the AI, tested the generated code, and gave feedback, and the actual work time was less than a few hours a day. 'It was like working in pairs with a very talented developer,' Warner said.

During this development process, Warner accidentally created a 'system' of four documents that summarized the architecture and workflow. This system consists of four Markdown files: 'pair_programming.md', 'project_plan_{some_extension}.md', 'technical_considerations.md', and 'mcp-browser-architecture.md'.

According to Warner, these files are the result of accumulated notes to prevent the AI from forgetting the context, and are intended to manage 'memory and forgetting' rather than the code itself. For example, 'mcp-browser-architecture.md' is a so-called README, which allows the AI to know the outline of the work even if it loses its memory. And 'technical_considerations.md' is for recording 'what problems do you not want to repeat?' However, these are merely the know-how that Warner gained during his work, and are not necessarily the correct answer.

From this experience, Warner believes that the skills required in modern software development are no longer knowledge of programming syntax, but the ability to convey 'precise imagination' and 'consistent aspirations' to AI.

Warner likens this to 'throwing spaghetti at the wall,' with one person responding, 'And then we see if it sticks,' to which Warner responded, 'It's not a question of whether it sticks. It's all about throwing it.'

Weird processes, failed experiments, and moments like 'it shouldn't work, but it just works' are all data points in the experiment we're running together without any hypotheses, Warner said. The above system is like spaghetti that happened to land on the wall in a recognizable pattern, and it might slide off the wall tomorrow, but Warner said, 'That's okay. We can just throw in some new spaghetti,' emphasizing the importance of trying things out without worrying about the results.

Warner sees the uncertainty of what lies ahead as scary, yet liberating and exciting at the same time, and says that for now he's enjoying building sandcastles through software development, knowing that the tide of progress will rise and engulf everything.

On the social news site Hacker News, there are comments such as, ' As natural language becomes the primary interface with technology, clearly expressing the challenge not only enhances communication, but also maximizes the potential of AI .' ' The key is to have a plan and be able to clearly express what you want. Then revise, adjust, and re-present it. You need to understand how software is built and how it works. If you don't have a good understanding of software architecture and product design, you can't expect the right software. ' ' Some people will lose their jobs, just like what happened in some industries when search became popular. But AI is just a tool. If you're smart, you can use it to make a lot of cool things. ' Some people say that engineers will be needed even if AI development becomes commonplace, while others say, ' I don't know what young developers will do (in this day and age when development using AI has become commonplace). They have to somehow jump to the top of the mountain, but the stairs are gone. ' There were also opinions that young engineers will lose the opportunity to gain experience.