Reflections Building Human-AI OS 4.1 by Ean Mikale, JD

#ai #blockchain #webholographic #nonagentic

In the past, we have never shared building reflections, as such. However, this build cycle we saw a profound shift, worthy of discussion. More than any other time in years, we hand-coded significantly, with a large reduction in AI-usage and code implementation. But why? That is the Trillion-dollar question.

Part of the answer is organization-specific, and the other aspect may be more generalizable across industry. The organization-specific aspect deals with our platform maturity, which has stabilized, and no longer requires significant changes. On the same end, platform stabilization allows us to more easily gain mastery over the codebase in front of us. On the other side of the discussion lies the increased constraints placed on AI models, and code generation. The models often provide large amounts of unnecessary, token-spending information. Many of the documents or resources produced are duplicative. Additionally, it is often faster to adapt code manually, than to wrangle with AI.

As a professional in the AI and Software industries, I hear about many of my colleagues who do not read the code AI generates. However, it is the routine scanning of code generation summaries, the codebase itself, and inline notations that we have found to be the most useful. In our experience, 9/10 times, if there is a persistent bug the AI cannot fix, it requires reading and understanding the code to: 1) Think of creative alternate solutions, and 2) Determine whether the code benefits from hand-coding over AI generation. Sometimes you might kick the vending machine (re-prompt), and other times the vending machine requires a knowledgeable repairman or repairwoman (who can read code) to get their hands dirty, and do things no one else feels comfortable, confident, or completely qualified to do.

Overall, where does this lead the AI industry? What are the broader implications? First, those initially learning software development, should NOT code with AI. Second, US Colleges and Universities have produced Computer Science degrees that do not produce functional programmers. Thus, the ability to read code, understand what is possible on the machine, and creatively problem-solve are more important than ever. In the near-future, such abilities contained in-house, will determine whether or not a company can maintain and scale AI-Assisted development. Professionals will also have to become self-starters, as the school systems are not architected to keep up with the pace of AI, or give you the hands-on experience necessary to land a software or engineering role.

In the broader context, this is a positive for humans, and for the workforce outlook. There are so many potential holes in the logic, and implementation of AI without a knowledgeable human-in-the-loop, that we foresee many downstream potential impacts, based on organizations pushing code without reading it. However, we also see potential for painstakingly patient companies. We have found that the details still matter. Craftmanship still matters. Passion and creative energy still matters. You, my friend, still matter.