Marvelous post, Dave. I wonder how much longer it will take Google and OpenAI to add LSP servers to their own agentic CLI solutions (codex and gemini).
On the solutions side, there is a very important aspect we should all watch for: even with models getting smarter, and with broader context, they are still a general purpose transformer.
Many general purpose products struggle to solve problems in a better way than specific solutions because of product canibalization.
In other words, they have to sacrifice a big part or tenet of its core product to become specific enough, and that often backfires.
For people investing in tooling, the future is not as gloom as it seems. At least for now.
These are decent takes. The business model (and technology) always follows the economics. Control is even more important to create those "moats" that are quickly being filled in.
On a personal note, I'm trying to control more of this by building my own personal lexicon of actions that'll kick off by just me typing them. As a developer it makes sense to maintain shell access and you're typical terminal experience but I thought it might be interesting to start building in other tools that I use often, like email.
But, it's the growing dictionary of terms, my own "LLM" but not really, I'm choosing every path that I want to make open (and stay open) and giving me total optionality. To a real degree, I've been architecting these types of systems for years but we now have better, formalized language. So, kind of like a Personalized Terminal Experience, or PTE.
Brilliant take on how LSP integration reshapes the whole tooling landscape. The key insight about integration moats vs intelligence moats is spot-on because enterprises will always need that messy auth and compliance layer that base models dont wanna touch. I've been working on similar semantic code analysis stuff and watched Claude just sorta absorb those capabilites each release. dunno if its better to pivot now or just keep building for today's needs instead of chasing teh moving target.
Marvelous post, Dave. I wonder how much longer it will take Google and OpenAI to add LSP servers to their own agentic CLI solutions (codex and gemini).
On the solutions side, there is a very important aspect we should all watch for: even with models getting smarter, and with broader context, they are still a general purpose transformer.
Many general purpose products struggle to solve problems in a better way than specific solutions because of product canibalization.
In other words, they have to sacrifice a big part or tenet of its core product to become specific enough, and that often backfires.
For people investing in tooling, the future is not as gloom as it seems. At least for now.
How to enable it?
These are decent takes. The business model (and technology) always follows the economics. Control is even more important to create those "moats" that are quickly being filled in.
On a personal note, I'm trying to control more of this by building my own personal lexicon of actions that'll kick off by just me typing them. As a developer it makes sense to maintain shell access and you're typical terminal experience but I thought it might be interesting to start building in other tools that I use often, like email.
But, it's the growing dictionary of terms, my own "LLM" but not really, I'm choosing every path that I want to make open (and stay open) and giving me total optionality. To a real degree, I've been architecting these types of systems for years but we now have better, formalized language. So, kind of like a Personalized Terminal Experience, or PTE.
Thanks for the good read!
Brilliant take on how LSP integration reshapes the whole tooling landscape. The key insight about integration moats vs intelligence moats is spot-on because enterprises will always need that messy auth and compliance layer that base models dont wanna touch. I've been working on similar semantic code analysis stuff and watched Claude just sorta absorb those capabilites each release. dunno if its better to pivot now or just keep building for today's needs instead of chasing teh moving target.
Enjoyed this one, thanks!