Am I more than a prompt? A developer’s existential crisis

Thank you, Claude Code, for my existential crisis.

For the last decade, my job has been in a niche combining my sports science background with software engineering skills. I was the developer that understood the domain, the sports scientist that knew how to build an app. It felt like a safe place to be. However, it looks like every sports scientist can now build an app. And any developer now has access to a seemingly unlimited knowledge base about training methodologies, power duration models, etc. Does my niche still exist?

LLMs and the developer tools like Claude Code that are built on top of them are wonderful pieces of technology. The speed at which you can now create boilerplate code, develop interactive mockups, create new features, write test cases and debug issues is incredible. And even domain-heavy development tasks are a breeze, having access to a seemingly unlimited knowledge base.

Seemingly unlimited. Not actually unlimited though. When you’re a mile deep in some exploration with your favorite AI model on a topic you really know a lot about, you bump into gaps in their knowledge, old paradigms or just plain errors. You only think you encounter these exclusively in conversations about topics you’re an expert at, but they’re actually present in all conversations, the “Gell-Mann amnesia effect”.

I’ve been working on a hierarchical categorization of sports for use in training applications (think “cycling”, “running”, but also “trail running”, “track running”). Ask Claude Code to come up with this and it will give you a very extensive and solid looking taxonomy with a lot of good things, but definitely also flaws: I ran a few tests and every time it had some weirdness in the grouping, for example grouping less popular “endurance sports” together while keeping cycling and running at the top level. An LLM doesn’t know what it’s like to search for your sport in a dropdown, or that it doesn’t make sense to combine training load from snowboarding and speed skating, just because they both are executed on frozen water. I do.

This is where I’ve started to rethink what my value actually is. Until recently, I looked at my new job as filling in the gaps in the LLM’s explicit knowledge, the knowledge that can be learned from reading and listening to people, or in the case of LLMs, training on vast amounts of data. But that’s a race you can’t win. There’s always an LLM out there that can write the code faster and better than you, or that has access to one more scientific paper about lactate kinetics. Maybe not now, but soon anyway.

So my value must be somewhere else: in tacit knowledge, the knowledge that can’t be taught or read, only experienced. Or, as Michael Polanyi originally said it: “we know more than we can tell”. The explicit knowledge gap closes. The tacit knowledge gap doesn’t.

This is not just about writing better prompts (“we know more than we can write in a prompt”). It’s about the direction the job of a developer is heading. The developers that will thrive are the ones that bring something to the table that the LLM doesn’t have. Not just better prompting skills, but domain knowledge and experience. The sports scientist that can code has always had an edge, but now it’s not the coding that’s the edge. It’s the sports science.

Since I started using Claude Code in September 2025, my daily work has changed significantly. Where before, about 80% of my time was spent writing and reviewing code (or fixing the bugs I introduced) and ~20% on planning, thinking about what to build, testing and other tasks, my current estimate is that this has dropped to about 20%. My job has changed into carefully iterating with Claude Code on new features, assessing whether its plan and generated code are correct and make sense. I went from planning-writing-iterating-testing to planning-iterating-testing. The writing was never really the job. The knowing was.

I am unmistakably much more productive than less than a year ago. LLMs have not taken my job, but definitely a significant chunk of it (and that’s a good thing). The niche isn’t disappearing. It just became harder to see from the outside.