Category shift · Market · Moat · Adjacent markets
The pitch in one paragraph: For 5 years (2020-2025), every product added AI features by adding text generation. JEV is the first model that returns decisions instead of text — typed, calibrated, schema-bounded. This is the same category shift as spreadsheets replacing ledgers: same data, fundamentally different interface. Quilt is the cellular substrate that turns those decisions into observable, signed, composable cells. The market is every B2B SaaS vertical that makes typed decisions.
LLMs (2020-2025) are text generators. Every product added "AI features" by integrating a chat API and showing the response. Total addressable: every human-facing software surface.
JEV (2026+) is a decision generator. The interface changes from "give me a paragraph" to "give me a typed answer with a probability." Every product can now add AI features by adding typed decision calls.
| Category | Before (text-based) | After (JEV-decision) |
|---|---|---|
| Search | User types query, system returns text | User types query, system returns top-3 docs + confidence |
| Forms | Regex validates format | JEV validates meaning |
| Routing | Rules engine with 5000 if-thens | 20 questions to JEV |
| CRM scoring | Lead-score formula in code | JEV calibrated score with explanation |
| Support tickets | Keyword match → team | JEV intent + urgency → team |
The spreadsheet analogy is precise: spreadsheets replaced ledgers because they made the cell the unit of addressable, composable, auditable value. JEV replaces LLM-prompt-as-decision because it makes the decision the unit of addressable, composable, auditable value.
Every B2B SaaS vertical. Lead scoring alone is a $4B market. Routing/triage is $8B. Form validation is $1.5B. Search relevance is $11B. These are decision-shaped problems.
The subset that today uses LLM tokens for typed decisions. At $25B, this is the achievable market with current JEV pricing and capabilities.
Apps that today burn LLM tokens on routing/validation/scoring because they have no cheaper alternative. Switching these to JEV is a 100x cost reduction.
| Volume | Per-call cost (JEV) | Per-call cost (GPT-4) | Monthly JEV bill | Monthly GPT-4 bill |
|---|---|---|---|---|
| 100k decisions | $0.0025 | $2.50 | $2.50 | $2,500 |
| 1M decisions | $0.025 | $2.50 | $25 | $25,000 |
| 10M decisions | $0.25 | $2.50 | $250 | $250,000 |
| 1B decisions | $25 | $2.50 | $2,500 | $25,000,000 |
1000x cost difference at high volume. This is the wedge.
First-mover moat. 6 months of inlined corpus + Workers + cells shipped live at ai-writings.pages.dev. Not a proof-of-concept — a production substrate with 115+ cell-being concepts, 82 inlined canon docs, 12+ live API endpoints.
Switching costs. Once your stack calls /api/jev/* for routing/validation/scoring, the entire decision layer is wired through this substrate. Replacing it means rewriting every decision point.
Data network effects. More decisions → better calibration → lower escalation rate → more decisions. The witness log gets denser. The model gets sharper. The substrate becomes more valuable the more it's used.
The cell layer as moat. Even if JEV commoditizes (TypeSafe open-sources, competitors catch up), the Quilt cell substrate does not. The cell algebra, the 11 opcodes, the witness log, the canon — that is the durable moat.
Decide without sending data to a US hyperscaler. Every healthcare, defense, and EU-regulated customer needs this. JEV runs on TypeSafe today, but the schema constraint makes edge-hosting tractable.
Every cell is signed. Every decision is justified by probabilities. Every witness is logged. Regulated industries (finance, healthcare, legal) can deploy AI agents because the audit trail is built in.
Merge decisions across distributed teams like Git merges code. Two people disagree about a routing rule? Run it through JEV. The decision log is conflict-free and reproducible.
A learning substrate where each cell is both student and teacher. A cell that gets corrected updates its own confidence distribution. The substrate learns from its own corrections.
Collective decisions about a community's data, made cheaply at the edge. A neighborhood can decide "share anonymized traffic data with the city" via a JEV-mediated vote, with each household's preference signed and traceable.