AI is making me anxious and stupid

I’ve been at a bit of a loss since the start of this year. Since Opus 4.5, LLMs including open weight models such as Kimi 2.5 and GLM 4.7 have (in my opinion) gotten incredibly good. While each release and improvement now is only incremental, the tooling and ecosystem around models is ever changing. The moat now isn’t who has the best model since they are all really good, but the tooling that you build around it (e.g. proprietary à la Claude Code or something FOSS like OpenCode). 

Keeping up with these developments is exhausting even if you’re  terminally online. Apparently MCPs are so last year and skills are the new thing? If you don’t have an AGENTS.md, cursor rules file, parallel subagents running in sandboxed Hetzner VPS instances using multiple models in a Ralph Wiggum loop with git worktrees that is split across different domains you’re not gonna make it bro /s. 

In all seriousness, I’ve been trying my best to keep up and give it a proper shot whilst still trying to “git gud” at things that traditionally make a solid engineer. People online will often talk about how important it is to have a solid understanding of the fundamentals so that you can drive LLMs correctly, and you don’t become overly reliant on them for everything that you build. 

But using AI feels genuinely addictive. It has become too easy to prompt an agent even for the easiest tasks because it is technically easier (although this is 50/50, since models still can fuck up in hilarious ways and end up taking twice as long). With an overly confident Claude or Codex, you feel as though you can learn anything and pick anything up instantly, and you begin to think that this must be true because that’s what all the cool kids on Twitter are saying. 

The problem is AI hasn’t always accelerated my work or learning. If anything, sometimes I unfortunately find myself trusting its take more than my own gut. Surely Anthropic and OpenAI with tens of billions of dollars in funding and some of the best researchers in the world have engineered this model to be smarter than me. I doubt myself, I think I’m a dumbass without Claude, I feel useless without an LLM and I need it to function as a developer. But this isn’t at all true, and if anything I am deliberately making myself (or at the very least making myself feel) stupider than what I actually am because I don’t really have that solid foundational understanding to fall back on. 

If anything, perhaps LLMs have highlighted the “Mr. Clumsy” nature in me. For context, Mr. Clumsy is a character from Roger Hargreaves excellent “Mr. Men” series of books, whose anxiety and nervousness is causing them to always doubt their own ability. Claude brings out the Mr. Clumsy in me as it accelerates my anxiety of “getting it right” the first time and not giving myself the chance to fail in a spectacular way. Mr. Clumsy gets frustrated at the feeling of being incapable of doing basic tasks, trusting your own ability to think, and that’s is maybe what AI has been doing to me. 

I think that I must not be the only Mr. Clumsy developer out there. The cycle is real, and I wouldn’t be surprised if others (particularly juniors to mid level like myself) felt the same way. On the one hand I’m comforted by people like Grady Booch who believe that similar events like this have happened before and AI won’t ever make us developers fully obsolete, but my ever growing dependence on these models is scaring the shit out of me because I am starting to doubt myself in terms of what I genuinely do know.

As a person and developer, I probably need to be more like Mr. Silly from Nonsenseland. In his case, he relentlessly pursuits the “Nonsense Cup”, and is a champion at keeping on. He just keeps on going regardless, much as others probably should despite what AI lab talking head is saying on Twitter or what my own internal Mr. Clumsy is saying. It will likely piss some people off by seeming overconfident and I would like to avoid this where possible, but I don’t want to become totally beholden to the shiny matmul machine. I would hope at many chances of failing but being able to smile as I did than begin to doubt in myself.