The Discipline
01 — How we build
We named our method before we sold it.
Every system we build draws on a library of engineering patterns — named, deliberately chosen, and validated against the published field: Anthropic and AWS on agent architecture, the security and compliance literature, the EU AI Act itself.
The library spans the whole stack, from the data substrate an agent reads to the audit trail it leaves behind — data patterns and AI patterns, from substrate to delivery.
ERP-agnostic by architecture — every system the agent touches connects through one governed tool layer. Live today against a mid-market European ERP.
The Method
02 — Assemble to act
Assemble
The agent gathers the evidence from wherever it lives: the ledger, the bank feed, the portal, the documents, the systems that don't talk to each other.
Propose
It proposes the entry, the classification, the reconciliation, the answer — grounded in the firm's own history and patterns, not a generic guess.
Check
A second agent reviews the proposal against the rules and the evidence before anything moves. Confidence is measured, not assumed.
Confirm
A person confirms. High-confidence, well-evidenced work moves quickly; anything uncertain is surfaced with its caveats for a human decision. Nothing is hidden.
Act — with a trail
Only then does the agent act: post to the system of record, file, write back — as an auditable action, attributable and reversible where it must be.
This is why our systems can do more than advise. They can carry the work — because every step is checkable, and the human stays in command of the parts that matter.
The Library
03 — The pattern library · data and AI, from substrate to delivery
The library, in the open.
Shared substrate, a data library, and an AI library — from substrate to delivery. The full pattern detail lives on the library page.
Shared Substrate
- Sovereign-by-architecture
- System-of-record preservation
- Cross-system entity resolution
- Provenance & lineage
The Data Library
- Data Substrate
- Ingestion & Movement
- Modelling & Semantics
- Quality & Governance
- Serving & Access
The AI Library
- Substrate
- Reasoning & Workflow
- Retrieval & Grounding
- Output Discipline
- Trust & Oversight
- Compliance & Delivery
The engineering is half the answer. The other half is control.