The Big Question in AI-driven developer experience: What's after Copilot?
Everyone is convinced this is "just" the beginning but can't articulate what the next killer form factor will be.
Increasingly convinced next step change won't be from optimal stuffing context windows
💡my take: Docs for humans, Embeddings for machines
Write samples -> get embeddings -> ship with each API/SDK version
✅ composable
✅ give users 99% accurate codegen
✅ can't version foundation models, but can version embeddings
If you liked @types, you'll *LOVE* @embeddings!
@swyx AI-driven code search would be great. I want to describe the functionality and have it return the modules and methods driving it.
Also, AI-driven CI tools that check new commits for bugs and unwanted behaviors.
@swyx well, "after" is a term that leaves a lot unspecified, but I wouldn't discount the value of things that can be built using *just* better prompt crafting / context building. There is so so much that is still possible there.
I'd also highlight that it's not all or nothing; embeddings can be used to improve selection of context for prompt. Hybrid approaches can probably achieve even more than the simple heuristics we are all sweating to discover today.
@swyx to elaborate a bit more: there are a ton of companies with on-prem requirements and not comfortable with sharing data to msft. Among them there are plenty who have custom DSL with their own data set to train the models on.
There is still no packaged solution (manage dataset, train model, deploy model, editor integration) that would support these self-hosted use cases.
swyx
Lex Toumbourou
Idan Gazit
Son Luong Ngoc