[simmonmt]

It jumped out at me too, but because I wondered what it would look like in the AI version of this story. Having had it build the SQL version do you ... a) miss the leap because you don't understand how it works, don't care to know, and go off to vibe the next thing b) ask it lots of questions because reasons to develop that deep understanding then make the leap or c) rely on it (prompt: "this can't be good enough do better") to go make the leap for you.

(Assuming for the sake of argument that you guided it to the SQL version first)

[embedding-shape]

Depends on what your overall goal is with what you're building. Is it to rush out as many features as you possible can, before VC-funding falls through the floor so you too can get a slice of the pie before the party is over? Or are you "retired" in your 30s and now have time to build the perfect software for you? Do you need to publish and release an experiment to see how people react to it or use it, before you can know if it's the right thing or not?

Almost everything needs to be contextualized before you can even begin to answer what the right way forward is, depends so heavily on what situation you're in.

[goosejuice]

Fwiw, giving opus 4.7 two sentences about building a cli doing Finnish to English translation and looking for a space efficient solution leads to an answer pointing to fst. For the same reasons stated in the blog. This is without a search tool.

The K shaped LLM scenario makes a lot of sense to me. Educated and experienced devs get better output because they know what to ask.