Post

I wrote up what I meant by this jeremymorrell.dev/blog/extensi...

So… my app should be like Salesforce or my app which is just a bundle of capabilities should ride on a platform like Salesforce? I read the thing. It lays out a pretty good account of how software that evolves to the level of abstraction that it does nothing. This treadmill seems to be a constant.

Salesforce has an architecture where it allows the user to add their own logic across many extension points. They had to build their own language and built their data logic into it. Today you have more options, but I think this architecture is a good fit for LLM-enabled users

This doesn't mean the software does nothing! The core provides the structure and guarantees that the extensions rely on. Think about Pi's core, or Obsidian's.

You can also take this to more or less extremes. Simple extensions like "when X happens I want to respond to it" that are basically a replacement for webhooks would also be super useful by themselves

Is “my app” in your imagination something like an e-commerce site or image sharing or something businessy like a CRM? I am probably just reading this from a weird perspective where all the apps I build are to solve a problem for myself and the apps I’ve worked for are like GitLab and honeycomb.

Turning my local apps into platforms is a huge waste but I’d love to have some rails to publish it where others could bring their own situations to it. Honeycomb should probably do a lot more custom edges. GitLab already had too many 6 years ago when I left.

Here's an example that I cut that might be closer to "turn local app into platform". Giving people the core structure of the thing so that the LLMs can write all the tedious glue code.

o11y and code hosting probably both have a lot of opportunity for easy extensibility here. There's so much in the gitlab UI that isn't relevant for me, and information I could bring in that likely is. My manager probably wants a completely different view of that same data.