- Personal user (not using in work)
- Auto-complete only
- Adopted agents, write specs + reviews code
- Adopted agents, write specs, no code review
- Adopted agents, no specs, no code review
- Moved back from agents to Auto-complete only
- Moved away from LLMs altogether
But the biggest value I have ever seen offered by individuals in the software industry comes in the form of institutional knowledge. The kind that allows some of us to hear about a few odd symptoms, think it through in our head, and immediately narrow down the problem space we need to troubleshoot into something small and manageable. Or can hear about a potential refactor and spit out a quick list of impacts it would have to the system in a few minutes on a call. Can LLMs do that? Maybe. Should they? I'm not so sure. If they can do in seconds what a human can do in days/weeks, that is clearly valuable. If they do in seconds what a human can do in minutes, that is less valuable. I lean towards a belief that human holders of the knowledge is a good thing, but LLMs could absolutely assist with knowledge transfer and sharing, making even more people able to work at that level. I'm simply unconvinced that outsourcing all that to LLMs is desirable.
I am 100% convinced that massive amounts of work, efforts, infrastructure build-out, and money could get LLMs to that point. But my question remains because that same level of resources could also do so many other things for our society. So I'm not stuck on "What could LLMs do?", I'm stuck on "What society do we want to live in?", and not at all convinced that the AI-pilled "AI can do it all" world is desired, even if it is possible.