One governed decision at a time
Enterprise Runtime ensures that AI-assisted work proceeds only from verified current enterprise state, under current organisational authority, with a result that can be explained against the state used for the decision.
AI Fabrix does not execute a model-generated plan or a predefined workflow.
A Role Assistant defines the governed business purpose and operating boundary. An approved AI interface can provide interaction. Enterprise Runtime owns execution; its Decision Engine owns each continuation decision. A continuation is the one next action Runtime permits: execute, wait, complete or stop safely.
The public Runtime model is simple:
Business purpose → current governed enterprise state → one governed decision → execute, wait, complete or stop safely → validated result → new governed state → decide again when required
This is the key difference from a conventional agent loop or workflow engine.
| Responsibility | Architectural owner |
|---|---|
| Governed business purpose and operating boundary | Role Assistant |
| Interaction and model reasoning | Approved AI interface / model |
| Execution | Enterprise Runtime |
| Each continuation decision | Decision Engine within Enterprise Runtime |
| Current authorized business facts | Enterprise Reality |
| Reusable organisational understanding | Enterprise Knowledge |
| Validated operational knowledge | Evidence |
| Canonical enterprise structure, compiled authority and capability information | COM |
| Current organisational authority enforcement | Operational Trust |
| Certified business capability execution | Enterprise Runtime through its governed execution boundary |
| Connected-system integration execution | CIP |
| Authoritative records, facts and transactions | Existing enterprise systems |
| Validated terminal business result | Process Execution Output |
Runtime uses these products and responsibilities. It does not transfer their ownership to a Role Assistant or AI model.
Enterprise Runtime evaluates a governed state snapshot used for one decision. It contains the information required to determine the next safe continuation.
That can include:
The snapshot is immutable for that decision. Runtime owns the governed execution state, and each validated Runtime result produces a new governed snapshot before another continuation is determined.
Authoritative execution state does not depend on hidden model memory or private agent state.
Because each decision is bound to a governed snapshot, the execution can be explained and replayed against the state on which that decision was made.
The Decision Engine evaluates the current governed snapshot and determines exactly one next continuation.
| Continuation | When it is used |
|---|---|
| Execute | Perform one bounded platform task, such as resolve current Reality, evaluate a capability or carry out an authorised enterprise operation. |
| Wait | Required information, authority or another dependency is missing. |
| Complete | The governed business objective has reached an accepted terminal result. |
| Safe stop | Continuation cannot be verified without guessing, using invalid state or bypassing authority. |
A useful sequence may emerge across many decisions, but no multi-step plan is committed in advance.
Enterprise Runtime uses several governed products without merging their responsibilities.
Enterprise Reality supplies current authorized business facts, including freshness and provenance.
Enterprise Knowledge supplies reusable governed organisational understanding and permission-aware knowledge retrieval.
Evidence supplies validated operational knowledge, obligations, decisions and business-significant conditions retained from governed work.
COM, AI Fabrix's enterprise compilation layer, supplies canonical enterprise structure, relationships, compiled datasource authority and certified capability information.
AI has no independent organisational authority.
Runtime evaluates governed work for a verified principal acting under an active Business Role and the organisation's existing authority model.
Operational Trust determines whether governed work is authorised to proceed. Where an enterprise operation is executed, that authority is revalidated against the current governed context immediately before execution.
Operational Trust evaluates the applicable identity, role, authority, policy, approval state, certification, data scope and permissions.
Prompts, retrieved content and model reasoning cannot create new organisational authority.
If required authority cannot be established, Runtime waits, denies the proposed path or stops safely according to the governed decision.
When the next continuation requires an enterprise operation, Runtime invokes a certified business capability rather than exposing raw vendor APIs to the AI.
Capability-level authority is enforced immediately before execution. The capability then uses Composable Integration Pipeline (CIP) for the governed connected-system operation.
Governed Runtime decision → Operational Trust → certified capability execution → CIP → authoritative enterprise system
CIP executes the integration. It does not own continuation, canonical enterprise meaning or organisational authority.
For current enterprise facts, Runtime obtains the result through Enterprise Reality's governed resolution path.
Read: Runtime determines Customer retrieval → Enterprise Reality resolves the authorized current fact → validated current fact returns to Runtime.
For a governed change, Runtime uses the certified capability path to reach the authoritative system.
Write: Runtime determines the business capability → Operational Trust and approval conditions are satisfied → certified capability execution invokes CIP → authoritative system accepts or rejects → actual execution result returns to Runtime.
A generic execution result is not automatically Reality.
A successful write does not automatically mean Runtime holds refreshed Enterprise Reality. Where confirmation of the new current state is required, Runtime resolves fresh Reality and makes the next decision against the updated governed snapshot.
This distinction keeps current business state fresh, authorized and attributable to its source.
AI Fabrix governs how AI-assisted work reaches enterprise systems. It does not replace those systems as the authority for the records, facts and transactions they own.
If an authoritative system rejects an operation, Runtime preserves that actual result. It does not convert the rejection into success because the model expected another outcome.
Some work requires information or explicit human authority.
An Ask requests missing information needed to continue. It does not grant authority.
An Approval records explicit human authority for the governed work and state for which it was requested. It does not create broader or permanent authority.
A validated human response becomes part of the governed Runtime context. It does not bypass Runtime or remove later trust checks. Material change requires authority to be evaluated again.
The Decision Engine evaluates the resulting new snapshot before determining what happens next. Until the required information or authority is established, Runtime waits or stops safely.
Safe stop is not a technical failure.
If Runtime cannot establish sufficient current context, authority or verified information to continue safely, it stops rather than guessing, using stale or conflicting state, leaking protected information or bypassing authority.
When continuation cannot be verified, it does not proceed.
The architecture is best understood as an ordered loop:
Enterprise operations inside that loop pass through Operational Trust and certified capability execution before CIP reaches the authoritative system.
This is the core Runtime law:
Enterprise Runtime does not execute a model-generated plan or predefined workflow; it makes one governed decision from current enterprise state, validates the result, and decides again.
A Process Execution is one governed attempt to achieve an accepted business objective.
When a Process Execution reaches a governed terminal outcome, Process Execution Output records the validated terminal business result and references the supporting governed results required by that outcome.
Conversation is interaction. Process Execution Output is the validated execution result. It does not automatically become Evidence.
Presentation through a Role Assistant, approved AI interface, review surface or application does not create a competing enterprise result.
A Runtime result does not automatically become Evidence.
Evidence retains validated operational knowledge only when the information is business-significant and belongs in the Evidence lifecycle.
Evidence can inform a candidate improvement, but it cannot silently become Runtime policy, capability or behaviour.
The architectural principle is simple:
Role Assistants define governed business purpose. Enterprise Runtime owns execution. The Decision Engine owns each continuation. Operational Trust protects authority. Systems of record remain authoritative.
Review the complete architecture in your own environment.
