tl;dr You still need a Software-Engineer-In-The-Loop to take the full advantage of LLM-enhanced (AI agent?) development. Personally, I’m not quite s

Daniel Olshansky on Substack: "#TIL About a Limitation of Claude Artifacts tl;dr You still need a Software-Engineer-In-The-Loop to take the full advantage of LLM-enhanced (AI agent?) development. Personally, I’m not quite sure when this will go away. One of the reasons I love Claude Artifacts [1] is because…"

submited by
Style Pass
2024-11-24 23:00:02

tl;dr You still need a Software-Engineer-In-The-Loop to take the full advantage of LLM-enhanced (AI agent?) development. Personally, I’m not quite sure when this will go away.

One of the reasons I love Claude Artifacts [1] is because it shows a company that’s building foundation models venture into the product space.

While working on a small project for my Dad (details to come in a full on post), I tried publishing a Claude Artifact but was hit with permissions issues. He was hit with a very annoying permissions issue:

Having spends many days hitting permissions issues in the past, you get some intuition for workarounds. In this specific case, I iterated until we got an HTML file we can download and open in a browser locally (i.e. no artifact needed).

This was very trivial, quick and easy, but required intuition that only came with years of software engineering experience. There’s no chance my dad, a non tech savvy civil engineer could have figured this out on his own.

Leave a Comment