9 posts
Formspec's Assist specification defines a standard protocol for AI agents, browser extensions, and accessibility tools to help people understand, complete, and reuse data across complex forms — without requiring an LLM.
Formspec's References and Ontology specs create a dual context layer — meaning plus evidence — that lets AI agents auto-fill, explain, and support form completion with grounded, domain-accurate answers instead of plausible-sounding hallucinations.
A federal grant application needs English, Spanish, and French — and a typo fix in the French translation must never re-trigger validation or require a new form version. Formspec's Locale Document separates translation from form logic entirely.
Formspec's new semantic layer gives form fields stable, machine-readable concept identity — the missing ingredient that turns AI data engineering from guesswork into something you can actually trust.
We decomposed XForms, SHACL, FHIR, JSON Forms, SurveyJS, and ODK into 517 testable features before writing Formspec. Here's what each system gets right, where each falls short, and what the gaps tell you about the state of form infrastructure.
Formspec was built as a chain of formal models — research into specs, specs into schemas, schemas into implementations. Here's how we did it in three weeks, and what we learned about AI-driven specification work.
We evaluated CEL, JSONLogic, JSONata, Power Fx, and JEXL before building FEL. Here's every alternative side by side — same expressions, six languages — and the Rust-based future that makes owning a language sustainable.
Why we built a new form specification, what it solves, and how it fits into the ecosystem of tools for grants, field operations, and compliance workflows.
Forms have been solved a hundred times — unless your data actually matters. A look at the gap between form builders and form infrastructure, and how six prior-art standards shaped Formspec.