About

jeesh starts from a simple bet: the cheapest way to run software and the best way to evolve it are the same move — climbing the stack of abstraction. Every layer humanity has added on top of raw machine instructions exists because someone found a way to say more by writing less. Programming languages are still the sharpest tool we have for that: a good abstraction doesn't just make code shorter, it makes the next piece of code cheaper to write, too.

jeesh leans on two properties that matter more, not less, in an AI-authored world. Declarative code says what should happen, not how — less of it to write, less to re-derive from scratch every session, less to disagree about. Interpreted code closes the loop between an idea and a running program to nearly nothing — no build step between "propose a change" and "prove it," which is exactly the loop an agent needs to run thousands of times a day, not a few times a release cycle.

Put those together and jeesh becomes a language whose entire standard library is a single, deduplicated wiki. Every rune is authored once, hashed, and shared. An agent that needs to add two numbers doesn't reinvent, test, and burn tokens proving add works — it looks it up, verified by hash, already proven by the wiki's own CI. Multiply that across every agent and every session, and the token savings compound the way code reuse always has — except the "reuse" is the language itself, not a library bolted onto it.

The other half of the bet is about who gets to change the language, and how fast. Most programming languages were designed for humans, by humans, at a specific moment — then they froze, because backward compatibility and decades of tooling investment make syntax nearly impossible to move once it ships. jeesh doesn't have that problem, because the language is the wiki: adding an operator or a construct is a content change, not an engine release. And because runtimes can (with consent) report which lookups failed, the language evolves from evidence of how it's actually used — a different clock than waiting for the next spec meeting.

jeesh is an early, experimental project conceived by thecapacity (Jay Huie). Right now there's a small, hand-curated vocabulary and a local CLI that runs it — see Getting Started. It isn't open to outside contribution yet, but that's the plan: once the guardrails are solid enough to trust a swarm of contributors, human and AI alike, pushing directly to the language, that door opens.