LLM driven development observations
In the LLM-driven development world, most of the engineering effort isn’t about solving the actual problem - it’s about wrangling the model like an overconfident intern who charges by the word, forgets everything you said five minutes ago, and still insists they’re right.
Your real job becomes:
- Prompt whispering - crafting polite, precise requests so the intern doesn’t hallucinate their way into JSON errors.
- Budget gymnastics - because every token costs money and your product roadmap didn’t include “optimize GPT billing.”
- Context juggling - rewriting history in fewer words so your intern doesn’t forget what planet they’re on.
All of this… just to get the model to maybe, possibly, hopefully return something close to what you actually needed.