Becoming claude-pilled

I was not really an AI person at all in my previous company (Cisco). We never really had a bar set, or expectations to be met when we shipped features / fixed bugs. In quite a few companies, the expectation now is to have x% of your code be AI-generated, which is a weird as fuck metric but I guess it pleases the overlords, and so it has become mainstream. But what mainly is interesting to me is not how much code AI generates, but how exactly we can review the code it generates efficiently.

Now that I’ve started using Augment at work and Claude code for personal use-cases, I’ve truly begun to realise how powerful these tools are to aid you, if not to fully replace a few parts of your daily life as a software engineer. Engineering is not merely prompting, but prompting and tooling for AI are now quite literally some of the most essential pieces of an ever-growing puzzle in the world of LLM generated content, and this holds true in particular for coding mainly because of how exhaustively these models have been trained on (almost) all of the world’s open-source code at least.

How much of a difference does it make?

Imo, not an insignificant amount. There is still work to be done as a developer, the code is not necessarily perfect right away. There will be flaws, a LOT of verbosity and then some. Testing will simply have a hundred extra cases no one really wants chalked out. But the fact that it reduces developer time is the main USP here.

Skills, skills skills

Claude skills and custom agents are the talk of town, but how useful are they really? It very much depends on how much you are willing to invest to set them up, and by invest I mean using literal brain power and not moolah.