Tom MacWright

2025@macwright.com

LLMs pivot to the aesthetics of thinking

There's a shift in the marketing of developer-oriented LLM companies toward thinking. The two latest examples are Cognition/Windsurf's Codemaps and Anthropic's Keep Thinking ad campaign. They also made hats that say thinking on them.

As Robin Sloan points out, they're diluting the words 'thinking' and 'reasoning.' I'd say that Cognition is starting to dilute the idea of 'understanding' too: their post posits that it shouldn't take 3-9 months to learn a new codebase and senior engineers shouldn't spend so much time helping other people learn.

I think we're on the cusp of the tech industry taking another glorious attempt at cracking the problems of teaching and education. It's interesting to see. It's also frustrating for anyone with critical thinking skills because we knew from the start that vibe-coded slop would be legacy code, code that nobody understands and nobody feels responsible for.

LLM-generated code explanations as a way of speeding up learning and removing that pesky responsibility to teach brings me back to The World Beyond Your Head, the book that I think about all the time. A quote:

Quite apart from the business appeal of MOOCs for universities (payroll is a lamentable thing), mechanizing instruction is appealing also because it fits with our ideal of epistemic self-responsibility.

One of the book's main arguments is that education is a transfer of culture as much as it is an exchange of knowledge. This starts with joint attention in childhood and continues on for the rest of your life: learning is a social activity.

But it's also funny to think about how we're trying to minimize the time required to learn things, as if it's another part of the industrial process that should be systematized and automated. In the future, learning might be like it is in The Matrix, where Neo gets kung-fu skills instantly zapped into his brain.

Honestly I think it could be interesting for LLMs to generate overviews of codebases, though they'll often be wrong. But as a way of truly understanding and being accountable for the code, it seems about as useful as a McKinsey consultant creating a slide deck for the executives: you get a semblance of understanding without the real thing. And people actually doing engineering probably need the real thing. I think that getting that, acquiring that real experience and real understanding, will continue to be joyously slow and arduous.