This is the text of the SpyneJS pre-seed deck, for readers and agents that do not run JavaScript. A Markdown version is at https://relevantcontext.io/spynejs/deck/deck.md.

AI-Native Application Structure

The language and grammar for AI agents building web applications.

Mission

Provide a new rules-based foundation for the web, optimized for AI innovation and human control.

The Problem

The Web is held back by the layer AI has changed the least: structure.

AI generates within structure — it doesn't invent it. Services and spec-driven processes outside the code can only patch code challenges, not resolve them.

  • Intent inconsistency. The same prompt generates a different application every time.
  • Outputs can't be verified. Generated code untrusted for critical applications.
  • No common rules for integration. AI output from databases, design systems, and external platforms only joins an application through hand-built wiring.

The Insight

At the core of all applications are three layers: View, Behavior, Logic.

Build a framework around VBL, and it can define application structure separately from its content: HTML, CSS, JavaScript.

SpyneJS goes a step further and separates that structure from content: View, Behavior, Logic, and the Content they carry — VBLC. Structure remains stable; content is replaced safely.

The Solution

A web stack for agents and developers to build & integrate VBL-structured applications.

Agents work with the rules and structure in the code. Developers and authors have tools to understand and collaborate.

What it unlocks:

  • Intent returns the same application structure, every time — deterministic, not guessed.
  • Code can be verified — output checked against the rules; trusted where it counts.
  • Everything speaks the same language — generated code from any tool, any agent, integrates cleanly.
  • Services assemble in, at any time — Assembly on the VBL pattern lets multiple services join the same application, at launch or years later.

A universal core: Every application shares the same structure, yet expresses the entire range of web experiences.

Why I Solved It

Frank Batista — Software Architect & Tech Lead

WPP · Discovery Channel · Morgan Stanley · Independent

I’ve spent years bringing ambitious visions to life — taking complex technical projects from initial design to successful launch.

I've built dozens of critical applications for well-known brands. The pain I experienced early on led to SpyneJS — the framework and system built on VBL. It streamlines every app into the same unified core, simplifying decision-making and the evolution of browser applications. SpyneJS enhanced my career, while providing the enterprise environments to bootstrap the open-source framework into maturity.

SpyneJS in enterprise production: a dozen-plus enterprise applications in use today, two external-facing — a Washington Post build from 2018 still runs untouched, and a Fortune 100 production site has evolved with the framework for six years, current release included.

AT&T · Accenture · BASF · Boeing · Cartoon Network · Chase · Citibank · Converse · Discovery Channel · Dunkin' · Emirates · Geico · Gillette · Hewlett-Packard · IBM · Metropolitan Opera · Microsoft · Nerf · Nuvigil · OneTouch · Pillsbury · Puppy Bowl · Scandinavian Airlines · Shark Week · Shell · Sony Pictures · SoulCycle · Toyota · United Way · Vonage · The Washington Post · and the estates of Otis Redding, Henry Mancini & Mavis Staples

Why Now

AI can now work with rules to generate structure.

The noise AI trained on can be replaced — with a compact set of rules that express the full range of web experiences.

AI code built on VBL structure unlocks capabilities that standard frameworks cannot deliver.

  • AI Code Assembly — complete applications from a plan, not guessed line by line.
  • Generated code from many services — every tool's output speaks the same structure; through plugins today, MCP next.
  • The ecosystem grows at AI speed — services and integrations built with the same agents the structure serves.
  • Developers build with their own agents — full collaboration and control, without the problems.

Proof

Four claims, verified — two by data, and two you can run right now.

VERIFIED: the noise is measurable, and SpyneJS removes it

The same application in Next.js and in SpyneJS, measured by noisemap, an open tool that classifies every token as View, Behavior, Logic or Content. Mean mixing fell from 0.24 to 0.003; consistency from 0.24 to 0.005. The parity result stands behind it: agents with no training on SpyneJS matched the Next.js app update for update. Open the Next.js map · SpyneJS map.

github.com/relevantcontext/benchmark-protocol

VERIFIED: the grammar replaces framework training data

Rule violations: from 344 to 19

With Generative Grammar

Same requests, with the grammar and without, output tested for correctness: mistakes against the framework's rules fell 344 → 19.

methodology in the repo

RUN IT YOURSELF: structure and content are assembled, instead of coded

npx spyne-cli create-app — applications expressed as data, assembled by lesser models: complete apps in under a minute, for under $0.02 each — measured. The running service, live today.

try it now

RUN IT YOURSELF: the CMS with AI exchange, installed in every app

Author content and move data between the CMS and any AI service — copy and paste, both directions. Working in every application today.

in every Application Shell

Traction

AI-native SpyneJS began in 2026.

  1. AWS-funded POC

    Worked with an AWS Rising Star partner, New Math Data, to determine what is structure and what is content.

    Spring 2026

  2. Generative Grammar

    Created from the success of the AWS project: the Generative Grammar — connecting AI's vast knowledge to VBL structure.

    Jul – Aug 2026

  3. AI Code Assembly

    Applications assembled — structure and content — by lesser models, the grammar already loaded. This deck was built with Assembly and the grammar.

    Aug – Sep 2026

  4. Public launch

    Launch demos and VBL education; begin pilot assembly with code generation services.

    Fall 2026

Assembly and the Generative Grammar evolve continuously. The grammar itself is open source — how we create it, and how we evolve it with AI, is proprietary.

Market

Web production, with an AI-native foundation.

  • The demand is for AI's full potential — 1,000+ AI companies ship frontends today, every one pushing against the current stack's limits: unpredictable output, unverifiable code, custom integration.
  • The pipeline is proven and ready — Assembly runs in production today, and the web production system gives teams the way to work with it.

Source: Mordor Intelligence, Web Development Services and CMS market reports (2026), global USD. SOM: ~1% of SAM, anchored to a bottom-up count of the first segment.

TAM $112B
SAM $31B
SOM ~$300M
  • TAM Global web production, 2025 — development services ($81B, to $134B by 2031) + CMS platforms ($31B)
  • SAM content-managed, hosted platforms
  • SOM ~1% of SAM, short-term — 1,000+ AI companies as the first segment

Business Model

Create the path to the new stack — and provide the premium tools to work with it.

  • Two tiers, one conversion: customers receive the code through partners — then subscribe to the Web Development & Authoring System: the CMS, Behavior Console, and CLI today; Workbench and Blueprint to follow. The subscription also opens the component library: the top 200 features and components, built in VBL, assembled at app creation and retrieved in the Workbench — cheaper in tokens than having agents rebuild them from the grammar every time.
  • Clean patterns replace noisy ones: the component library is built as VBL features — functionality, data integration, and presentation kept apart and combined by rule. Where models today match against a decade of mixed code, agents on SpyneJS match against units that were built clean, and mix and match cleanly.
  • Adoption pulls partners in: the grammar already works with the agent tools services ship today — no integration required. As a service sees its output land cleanly in SpyneJS, we build a custom grammar with them: 20–40 hours, not a months-long integration.
  • Revenue: Web Development & Authoring System — recurring subscriptions, with the component library in the paid tier. Conversion service — paid engagements that migrate existing codebases onto VBL, scoped with the noisemap. Partners — custom grammars and licensing.
  • First customers: greenfield enterprise teams that need verifiable code — and strong category competitors looking for an edge.
  • Partner selection: polished UX · structured output · strategic need to challenge a leader · distribution to application users. Initial reference path: design → builder → backend → one SpyneJS application.

Competition

AI-generated code assembly

The current stack

Generated-code APIs are under exploration — Vercel and Google among them. Each unifies what a generator hands to a runtime; beyond that, integration is one library per service, with the structure between them left to the developer. Every attempt we found unifies at the API; none unifies in the codebase.

SpyneJS

VBL conforms generated code and web services to one shape. Every source assembles into a single codebase — one structure, readable end to end. The APIs under exploration are useful inputs: a streamlined path from generation services into VBL.

Web production

The current stack

Next.js at the framework tier, and WordPress- and AEM-class systems at the platform tier, hold deep integration and large user bases. They are betting that faster, better AI models will minimize the pain of AI's problems: inconsistent intent, unverifiable output, no common rules.

SpyneJS

The platform is built on what faster, better models alone do not supply: precise intent, verifiable output, and legible code.

Merging generative AI, design systems, and web services into a unified codebase multiplies their value — each source compounds the rest.

The moat

Roadmap & GTM

Driving adoption through agents.

Who we target: enterprise teams that need code generation for critical applications — where verifiability is the requirement, not a feature. And the users of partner services: every partner-generated application creates a SpyneJS user, acquired by the partner, converted by the Workbench.

Two tiers, two motions:

  • Partners — deal-shaped, founder-led: a handful of targets chosen by the selection criteria; their wins co-marketed.
  • Production system — pull: launch, VBL education, and tutorials across Assembly, CMS, and Workbench — converting the users partner applications create.

The economics of acquisition: distribution concentrates in a handful of partner deals — not thousands of acquisitions — and end users arrive through partners at no marginal cost, and a custom grammar takes 20–40 hours, not a bespoke integration.

Phased: 0–6 months — first partner launches: grammar proven with services' own agent tools, first custom grammars, first paid partner engagements · 6–12 — prove the network: three categories, 3–5 partners, registry open, multi-partner combinations · 12–24 — launch and scale: 6–8 partners, 25+ production applications, 10+ paying organizations.

Fall 2026: public launch → first partner conversion → Workbench live

  1. Partners' agents assemble on SpyneJS

  2. Applications ship to their users

  3. Users convert in the Workbench

  4. The platform pulls the next partners

The Ask

$1.5M–$2M turns the assembly engine into a repeatable partner business, and a web production system built for this time.

$1.5M–$2M · Pre-seed (SAFE), rolling close · 18–24 months runway

~70%
~20%
~10%
  • Product & engineering
  • Partner enablement & GTM
  • Operations

Milestones funded, not headcount.

What this funding delivers:

  • The production system, launched — Workbench and Blueprint join the CMS, with full admin and authoring: a direct competitor to WordPress and AEM.
  • Multiple strategic partners live — 6–8 contracted, shipping working assembly output on SpyneJS; 25+ production applications, 10+ paying organizations.
  • Assembly and the Generative Grammar, grown — deeper coverage, continuously evolved with AI.
  • The GTM engine running — launch, devrel, and the publishing program across platforms.

Each milestone positions the seed round: a shipped competitor product, revenue-bearing partners, and a repeatable partner playbook.

Thank you.

frankbatista@relevantcontext.io · spynejs.com