Has Anyone Here Tried Creating Their Own Programming Language?

Been deep in an interesting rabbit hole lately — designing a non-branch programming language called Zeus inspired by Konrad Zuse's Plankalkül from 1945. The concept is that instead of writing if/else and loops, all decisions are encoded as data selections and all iterations become map transformations.
What makes it interesting from a money perspective is that it could become a universal intermediate representation — meaning Python, JavaScript, any language could compile into Zeus graphs and back out again. More importantly it has potential as an internal reasoning layer for AI models, making AI decisions fully auditable and traceable instead of the black box they currently are.
Anyone here experimented with building DSLs (domain specific languages) or custom scripting languages for their tools or automation? Curious what approaches people have taken and whether there is a market for language level tooling in the internet marketing space.

Youngbill44 said:

all decisions are encoded as data selections and all iterations become map transformations

Click to expand...

This idea was popular around 8 years ago, when Kotlin, Swift, Scala became popular

NomixGuy said:

This idea was popular around 8 years ago, when Kotlin, Swift, Scala became popular

Click to expand...

Yea but u don't understand this may create a new era of prompting ai using the prompt can create accuracy on ai to create the same thing it doesn't matter the ai am still experimenting with it

Try and make something better than C then? lol Unless you write direct from assembly EVERYTHING has C as its parent language

Youngbill44 said:

Yea but u don't understand this may create a new era of prompting ai using the prompt can create accuracy on ai to create the same thing it doesn't matter the ai am still experimenting with it

Click to expand...

Honestly didnt understand what you're saying bro. Era of creating programming languages has past away. Check latest news, llms invent their own languages already, even tho nobody asked them lol.

Indeed but ur wrong actually programming is creation of everything ur seeing today

@Youngbill44 tbh most guys in the marketing space won't care about a new language... they just want a tool that prints money or automates their setup. We used to build basic DSLs for custom scraping tools years ago just so virtual assistants could write simple configs, but python and node pretty much took over everything.

The deterministic AI reasoning layer you mentioned is the only part that might have legs. If you can use this to stop LLMs from hallucinating and make them follow strict workflows without massive API costs, there is definitely a market. Just don't pitch it as a programming language to marketers or they will tune out immediately. Pitch it as a way to make AI agents actually reliable.

@The WP Nerd nailed the real angle imo. Nobody in marketing buys a language, they buy reliability. The "auditable AI reasoning" framing is the only thing here with money behind it.

@Panther28 the C parent thing isnt really true tho... Lisp, Fortran, ML, Prolog none of those came from C and a lot of modern stuff traces back to them instead. Plankalkül itself predates C by like 25 years lol. Not trying to be that guy just sayin the lineage is messier than that.

On the actual idea Youngbill... the map/select instead of branch thing is basically what dataflow langs and stuff like Haskell already lean into, and IRs like LLVM or MLIR already do the "compile everything into one graph" job. So your hard part isnt the concept, its convincing anyone to add another layer when those exist. The reasoning-trace use case is genuinely different though, that space is wide open right now and people are throwing money at anything that makes agents less of a coinflip.

Build a tiny demo where an agent runs through your graph and you can replay every decision step by step. Forget the marketers, take that to the AI tooling crowd. Thats where it has a shot.

Nexora550 said:

Lisp, Fortran, ML, Prolog none of those came from C and a lot of modern stuff traces back to them instead.

Click to expand...

Your right! Never learn by memes! lol

the auditable reasoning angle is the one worth chasing but i'd push back a little on "no API costs". if your graph still calls an llm at each decision node you're paying for every step, the savings only show up if you can cache or short circuit the deterministic parts. thats where the actual engineering is.

honestly the replay demo idea someone mentioned is the move. dont even talk about the language. build a thing where an agent does a multi step task and you can scrub back through every choice it made and see why. that alone gets attention from the agent crowd right now cause everyone's terrified of agents going off the rails in prod and nobody can explain what happened after the fact.

one thing tho... compiling python/js into your IR and back out is a massive undertaking and tbh i dont think it adds anything to the money story. LLVM and MLIR already exist and you'd be competing with people who've spent a decade on it. id drop that part entirely and just focus on the trace layer. narrow it down or you'll spend two years building infra nobody asked for.

what does your decision node actually look like right now, are you encoding the branch as a lookup table or something fancier?

@Fluxify is right about narrowing it down. One thing nobody really said tho... the people who'll actually pay for an audit trail arent the agent hackers, its compliance. anyone running agents anywhere near money or health data needs to prove why a decision got made after something blows up, and right now they got basically nothing. if your graph can spit out a tamper evident log of every select step, thats a deliverable you can sell on its own without ever calling it a language.

and yeah drop the compile python/js in and back out part. you'll burn a year fighting LLVM/MLIR people for no payoff. easier move is just instrument an existing agent framework so it emits your graph, dont make anyone switch their stack to try it. meet them where they already are. once a few people see a clean replay they'll come asking how it works.