Ben Jacobson

CTO at Onboarded - AI Native Compliance Infrastructure for HR

[colinhacks/zod @colinhacks]

going nuts because no AI app lets you have separate windows per repo/workspace (y'know, like IDEs did for 30 years)

plz fix @ claude codex conductor cursor

[Ben Jacobson @bjacobso]

@colinhacks I asked conductor early about this, didn't get a great response. @superset_sh said they would be working on it eventually. Desktop software must support all desktop primitives.

[Ben Jacobson @bjacobso]

@coetry @ctatedev All languages will slowly converge to lisp, and all agent tools will slowly converge to a lisp repl

[Ben Jacobson @bjacobso]

@coetry @ctatedev This is exactly what I have been working on, would love your feedback https://t.co/AQHuH7Holu

[Ben Jacobson @bjacobso]

My thoughts are still evolving, and I do appreciate how crazy this sounds: it started a few months ago with an idea called Open Ontology, my goal was to build a Palantir-like way to model operational ontologies. This turned into a triplestore, kind of similar to event sourcing, but on a attribute/entity/transaction level. I got that working as an Effect library, pluggable storage (sqlite/durable objects/pg/fndb/etc). I then needed a query engine, so I built a datalog engine on top of that triplestore. Then I needed a way to model the actions and mutations and constraints. Then compose those into workflows, then allow users to define views, etc. at some point I realized I wasn't really building a datastore anymore, I was trying to model the operational semantics of an organization, and every layer kept wanting to become declarative (because I am a BIG schema fan). I wanted workflows to be data, then actions to be data, then permissions to be data, then views to be data, then eventually logic itself wanted to be data, just like HttpApi. I could do this with Schema, but if the goal was to embed in markdown, why limit to typescript? The goal is also not authoriing ergonomics, its reading ergonomics. This culminated in the idea of embedded a lisp dsl into markdown, allowing a markdown file to function as both the spec as well as the implementation. Lisp serving as a harness for highly concise dsls. However, it needed HM type checking and built in support for effect and schema modeling. So I set down the path of building a custom lisp to support this concept. I'm currently calling this language Forma. In the end I have the whole system mostly working including LSP and IDE support, terraform style deployments with git / PR like ux, and a runtime to actually run the entire graph. Because it is lisp, it is homoiconic, everything is completely auditable, the data, the actions, the schema, the execution, and the evolution. I can argue Schema is also a Lisp, and could get all the same benefits, but the lisp syntax turned out really clean. I haven't open sourced it yet, because that is a one way door and I don't have a lot of open source experience. Some domains I am playing with:

https://t.co/0pql3Lgd1p (thinking a way to share ontology definitions- Employees, Departments, Projects, etc. as a public library of operational ontologies, or modules like Stipe of Salesforce)

https://t.co/NqUlf7UPem (maybe a hosting service for the above, but not sure yet)

https://t.co/7zCbUTeDl0 (the lisp dsl for executable declarations, with built in effect and schema modeling, and an elaboration phase for designing custom dsls)

https://t.co/fJbjXqYYpN (my attempt to put a name to the primitive, the company behind the above, maybe a consulting / implementation shop for organizations that want to use this technology, tbd)

Again, I know this is crazy, and very much an LLM induced psychosis, but I think LLMs should make our ambitions with software much bigger, and this is mine

[Ben Jacobson @bjacobso]

@dillon_mulroy @alchemy_run any chance I could get access?

[Ben Jacobson @bjacobso]

@zeeg And once you create a PR it’s not obvious it’s merged or if you have uncommitted or unpushed work

[Ben Jacobson @bjacobso]

@AndaristRake @MichaelArnaldi @tim_smart @EffectTS_ @ahejlsberg We should write down the sequence of events for the eventual documentary 🙌

[Ben Jacobson @bjacobso]

@campedersen makes sense to me

[Ben Jacobson @bjacobso]

Yo dawg

[Ben Jacobson @bjacobso]

My thoughts are still evolving, and I do appreciate how crazy this sounds: it started a few months ago with an idea called Open Ontology, my goal was to build a Palantir-like way to model operational ontologies. This turned into a triplestore, kind of similar to event sourcing, but on a attribute/entity/transaction level. I got that working as an Effect library, pluggable storage (sqlite/durable objects/pg/fndb/etc). I then needed a query engine, so I built a datalog engine on top of that triplestore. Then I needed a way to model the actions and mutations and constraints. Then compose those into workflows, then allow users to define views, etc. at some point I realized I wasn't really building a datastore anymore, I was trying to model the operational semantics of an organization, and every layer kept wanting to become declarative (because I am a BIG schema fan). I wanted workflows to be data, then actions to be data, then permissions to be data, then views to be data, then eventually logic itself wanted to be data, just like HttpApi. I could do this with Schema, but if the goal was to embed in markdown, why limit to typescript? The goal is also not authoriing ergonomics, its reading ergonomics. This culminated in the idea of embedded a lisp dsl into markdown, allowing a markdown file to function as both the spec as well as the implementation. Lisp serving as a harness for highly concise dsls. However, it needed HM type checking and built in support for effect and schema modeling. So I set down the path of building a custom lisp to support this concept. I'm currently calling this language Forma. In the end I have the whole system mostly working including LSP and IDE support, terraform style deployments with git / PR like ux, and a runtime to actually run the entire graph. Because it is lisp, it is homoiconic, everything is completely auditable, the data, the actions, the schema, the execution, and the evolution. I can argue Schema is also a Lisp, and could get all the same benefits, but the lisp syntax turned out really clean. I haven't open sourced it yet, because that is a one way door and I don't have a lot of open source experience. Some domains I am playing with:

https://t.co/0pql3Lgd1p (thinking a way to share ontology definitions- Employees, Departments, Projects, etc. as a public library of operational ontologies, or modules like Stipe of Salesforce)

https://t.co/NqUlf7UPem (maybe a hosting service for the above, but not sure yet)

https://t.co/7zCbUTeDl0 (the lisp dsl for executable declarations, with built in effect and schema modeling, and an elaboration phase for designing custom dsls)

https://t.co/fJbjXqYYpN (my attempt to put a name to the primitive, the company behind the above, maybe a consulting / implementation shop for organizations that want to use this technology, tbd)

Again, I know this is crazy, and very much an LLM induced psychosis, but I think LLMs should make our ambitions with software much bigger, and this is mine

[Ben Jacobson @bjacobso]

@Eigengrau001 @teej_dv sure, but think of ai as a tool that allows you to solve bigger problems than you could before. like a calculator or an ide.

[Ben Jacobson @bjacobso]

@DavidKPiano I'm working on this! https://t.co/AQHuH7Holu

[Ben Jacobson @bjacobso]

My thoughts are still evolving, and I do appreciate how crazy this sounds: it started a few months ago with an idea called Open Ontology, my goal was to build a Palantir-like way to model operational ontologies. This turned into a triplestore, kind of similar to event sourcing, but on a attribute/entity/transaction level. I got that working as an Effect library, pluggable storage (sqlite/durable objects/pg/fndb/etc). I then needed a query engine, so I built a datalog engine on top of that triplestore. Then I needed a way to model the actions and mutations and constraints. Then compose those into workflows, then allow users to define views, etc. at some point I realized I wasn't really building a datastore anymore, I was trying to model the operational semantics of an organization, and every layer kept wanting to become declarative (because I am a BIG schema fan). I wanted workflows to be data, then actions to be data, then permissions to be data, then views to be data, then eventually logic itself wanted to be data, just like HttpApi. I could do this with Schema, but if the goal was to embed in markdown, why limit to typescript? The goal is also not authoriing ergonomics, its reading ergonomics. This culminated in the idea of embedded a lisp dsl into markdown, allowing a markdown file to function as both the spec as well as the implementation. Lisp serving as a harness for highly concise dsls. However, it needed HM type checking and built in support for effect and schema modeling. So I set down the path of building a custom lisp to support this concept. I'm currently calling this language Forma. In the end I have the whole system mostly working including LSP and IDE support, terraform style deployments with git / PR like ux, and a runtime to actually run the entire graph. Because it is lisp, it is homoiconic, everything is completely auditable, the data, the actions, the schema, the execution, and the evolution. I can argue Schema is also a Lisp, and could get all the same benefits, but the lisp syntax turned out really clean. I haven't open sourced it yet, because that is a one way door and I don't have a lot of open source experience. Some domains I am playing with:

https://t.co/0pql3Lgd1p (thinking a way to share ontology definitions- Employees, Departments, Projects, etc. as a public library of operational ontologies, or modules like Stipe of Salesforce)

https://t.co/NqUlf7UPem (maybe a hosting service for the above, but not sure yet)

https://t.co/7zCbUTeDl0 (the lisp dsl for executable declarations, with built in effect and schema modeling, and an elaboration phase for designing custom dsls)

https://t.co/fJbjXqYYpN (my attempt to put a name to the primitive, the company behind the above, maybe a consulting / implementation shop for organizations that want to use this technology, tbd)

Again, I know this is crazy, and very much an LLM induced psychosis, but I think LLMs should make our ambitions with software much bigger, and this is mine

[Ben Jacobson @bjacobso]

@Eigengrau001 @teej_dv saying you aren't into AI is like saying you aren't into linear algebra. It doesnt make linear algebra just disappear.

[Ben Jacobson @bjacobso]

@Eigengrau001 @teej_dv it is philosophical. Any agent needs a shared ontology and definition to operate on any problem at any scale. Palantir started building solely for human operators, they are just benefiting from the AI wave today because it turns out The Ontology is valuable there too

[Ben Jacobson @bjacobso]

@Eigengrau001 @teej_dv https://t.co/A2zrRy2BGT started as an exploration of a small scale version of that pitch, and evolved into a new lisp language and triplestore and markdown based editing surface / ide

[Ben Jacobson @bjacobso]

@Eigengrau001 @teej_dv this is just my educated guess. I'm more referring to their mental model for re: The Ontology

[Ben Jacobson @bjacobso]

@teej_dv Similar to like json-render or foldkit etc

[Ben Jacobson @bjacobso]

@teej_dv This is functional today. Including a view layer, which eventually can be exposed as an mcp-app or something
https://t.co/hHKjhgnuC1

[Ben Jacobson @bjacobso]

@teej_dv initial goal is just possible to model, but eventually share models. for example Stripe or ADP or Salesforce etc don't need to be remodel by everyone, although vendoring those models should also be supported