Two sources of truth
Product managers write specs. Engineers invent schemas. Ops invent exceptions. Support invents workarounds. None of those artifacts are the running system — so drift is guaranteed.
Why
The gap is the product.
Every growing company pays a tax: the business changes in language and process, while the software lags behind in tickets and rewrites. That gap is not a documentation problem. It is an architecture problem.
Low-code still hard-codes the business
Forms and workflows help until the domain gets serious: versioned entities, invariants, delegated permissions, projections, login providers. Then you are back to custom engineering — without a clean model of what “correct” means.
AI makes the wrong layer faster
Generating more application code faster does not fix divergence; it accelerates it. What you want AI to accelerate is the specification — and a runtime that will not execute an inconsistent one.
Our bet
The durable artifact of a company is not a stack of microservices. It is a precise description of records, rules, and operations — compiled and enforced. Software should be a consequence of that description, not a parallel invention.
If the model is wrong, the engine should refuse. If the model is right, the engine should not need a rewrite.
What changes when the model runs
- New domains stop being new apps Products, inventory, access, onboarding — modules and definitions instead of greenfield CRUD each time.
- Rules stay next to the change Invariants and pre-action checks are definitions, reviewed like product, executed like code.
- Agents have a contract An LLM can propose a definition change. The compiler and the engine decide whether the world still makes sense.