Why Your AI Needs a Fishplate (and Why OpenAPI is Dead)
The metaphor of derailment
In the early days of the Industrial Revolution, the biggest threat to progress wasn't the engine's power—it was the alignment of the rails. If two rails diverged by even a fraction of an inch, the result was a catastrophe. To solve this, engineers created the fishplate: a structural joint that bolts two rails together, ensuring they stay in perfect, continuous alignment.
In 2026, we are facing a similar crisis in software. We have the engines (LLMs like Claude and GPT), but our rails—infrastructure and code—are buckling. We call this architectural drift.
The token tax of legacy specs
For a decade, OpenAPI (Swagger) was the gold standard. But OpenAPI was never designed for an AI-native world.
- Bloat: A standard spec can consume 50k tokens. That's a token tax on every prompt.
- Lack of intent: OpenAPI tells an AI how to call an endpoint, but not why it exists or what the NIST 800-53 constraints are.
- Static reality: OpenAPI is a promise, not a proof. It doesn't know if your Cloudflare D1 database actually matches the spec.
Enter LAPIS: the token-minified contract
At fishplate.dev, we've replaced the bloat with LAPIS. It provides the AI with a mental model of your system that is 85% smaller than OpenAPI, while being more rigorous for operations. It includes your Drizzle schemas, your Cloudflare bindings, and your GDPR sovereignty rules in a single, versioned Fishplate structure.
The age of prompt and pray is over. It's time to gird your systems.