Leandocs
An agent-native documentation workflow that generates bounded discovery surfaces and benchmarks repeatable API tasks for success, cost and failure quality.
API companies increasingly expose documentation to software agents as well as humans. Leandocs proposes a supply-side workflow: ingest an approved interface contract and repository scope, generate a constrained agent-tool surface and machine-readable documentation index, then benchmark representative tasks before and after publication. The supplied research confirms that documentation platforms already generate agent-tool servers and code-derived docs, while finding no product that publishes cross-API task-completion token cost as a measured specification.
Token count alone is a poor scoreboard. Interface version, documentation build, task definition, fixture, credentials, model and configuration, run seed, tool calls, token usage, latency, completion result, policy violation, evaluator decision, publication approval and customer outcome must remain separate. A cheaper run that returns the wrong object or overreaches authorization is not efficient. Benchmark claims need repeated trials, uncertainty and reproducible public methodology.
Generated tool servers expand security surface. Authentication, tenant scope, destructive actions, rate limits, error semantics and secret handling require owner review and tests before publication. One referenced capability remains unverified. Numeric competitor prices are omitted because they are observed market references, not fixed product pricing.
A developer-experience, API product or documentation leader at an API company competing for successful human and agent integrations.
Developer-experience and API product leaders own documentation quality and integration success.
A public efficiency score encourages better docs while incentivizing vendors to optimize narrow token metrics rather than safe task success.
Agent usage creates a new measurement target, but no strong historical barrier prevented benchmark tooling.
The input identifies a clear API-company buyer, confirmed agent-documentation generation and an unoccupied task-cost benchmarking layer that can become a shared scoreboard.
Generation is replicable, one capability remains unverified, benchmark validity is expensive to maintain and incumbent documentation platforms can add measurement.
Discussion
No comments yet — be the first to weigh in.