Agentic Sustainability

Who maintains a literate language? Not hired developers — they are expensive, and languages built for reading rarely excite them. The bet behind WikiScript is that agentic coding changes the answer: a decentralised commons of passionate volunteers, each directing AI agents, can carry a language no firm would fund. This page names the bet and the infrastructure it needs.

# The evidence of the shelf The passion has always been there; the engineering was the bottleneck. HyperTalk's fans outlived HyperCard by decades, and OpenXTalk volunteers still keep a 2021 engine alive. SimpleTalk was two people and a lab. Lexon is essentially one author. Every one of these ran ahead of what unfunded hands could sustain — not ahead of what people cared about.

# What agents change A well shaped codebase lets a volunteer's agent do in an evening what once took a contractor's month. SimpleTalk's whole language core fits in a single context window; the WikiScript kernel is smaller still. The economics flip: the binding constraint stops being the size of the team and becomes the shape of the project — how safely a stranger's agent can contribute without breaking what matters.

# The shape that invites agents - a small deterministic kernel — nothing to break subtly - the corpus as conformance — golden permission tables every change must preserve - the grammar as a document — an `.ohm` file fans can read, not only compiler people - typed seams — components with `process` interfaces, page-json in and ghost page-json out - wiki deliberation — design happens on pages anyone can fork - Radicle replication — no platform to be pulled away; patches travel between peers, per Radicle Pi Plan - signed contributions — the WikiScript security model decides whose agent's work you run

# Why literate languages especially The corpus is readable by the people who care — mediators, organisers, teachers. They can author agreements and the tables those agreements should yield, which is to say they can write tests, without ever writing a compiler. Agents do the compiler work; the fans hold the taste. In most languages the community that loves it and the community that can build it barely overlap. Here they are the same people, amplified.

# The risks Drift, first: agents produce plausible wrong changes at volume, and only the conformance suite stands in the way — which is why the corpus is the constitution. Governance, second: who merges, who signs, whose trust counts — the trust graph question, not solved here. And the SimpleTalk lesson: a bus factor of two put a fine codebase to sleep. A commons needs more doors in than it has founders.