Guardrails
I nearly shut down agiledrillsergeant.com at the start of the year. As it was coming up for renewal, I thought, why bother? I haven’t posted a comic in 3 years. I don’t really work as an agile coach anymore. It seemed to belong to a previous phase of my life. But because I can procrastinate past the point of action, the site got renewed and stayed available. What the hell, I thought, maybe there’s some agile coaches still left who can take some enjoyment from it.
Then, like my people I suspect, my work life got steam-rolled by AI around about February/March. I don’t mean by that, that I first heard of AI then, or that I looked at it for the first time. It was the first time, that it seemed that writing code with AI may actually be an economical option. It became clear that enough progress had been made that AI was not going away. That all my experience in software development would be a museum piece in 6 months time. At least that’s what it felt like!
I think the picture has become more nuanced since then.
I can see how you could create enough guardrails for AI to write usable code. However, that’s the short-term picture. These guardrails need to be maintained – and my theory is that to be cable to maintain the guardrails, you’ll still need to be able to program. You’ll have to still get your fingers dirty. I can imagine that certain issues around performance or architecture will not be solved optimally by an LLM trained on the median of GitHub. The devil is in the details.
I know there will be talk about feedback from the guardrails, self-correcting AI, production ready enterprise grade code. I just don’t buy it. It’s utopian – and no utopia has ever come to pass. AI will change how we develop software – just not the way the extremists imagine!