What is the ISE Framework?
The Interconnected Systems Environment (ISE) is a formal, graph-based framework for modelling how AI outputs propagate through enterprise systems. Introduced by Myriam Ayada (2026), it represents an organisation's AI-integrated infrastructure as a directed graph G = (V, E) where each node is a system and each edge is a corridor: a boundary where outputs transform into inputs for the next system.
Key insight. Certain failure modes in AI-integrated environments arise not within systems, but at the boundaries between them. ISE makes these boundary transformations explicit and analysable.
Why Model at the Boundary Level?
Current AI monitoring watches individual systems: model accuracy, data quality, component metrics. This is necessary but insufficient. When AI outputs cross system boundaries, meaning changes.
For example, consider a credit scoring pipeline. An applicant with a risk score of 0.39 vs 0.37 may be indistinguishable to the AI. However to a categorisation engine with a threshold at 0.38, one produces MEDIUM and the other LOW. ISE shifts attention from "is the model right?" to "what happens to the output after it leaves the model?"
Sample ISE pipeline
- Node 1 — AI Risk Scorer (Semantically open, AI-source) → discretises →
- Node 2 — Categoriser (Rules engine, spec-closed) → routes →
- Node 3 — Decision Engine (Approve / Deny / Review, spec-closed) → aggregates →
- Node 4 — Calibration Module (Portfolio monitoring, spec-closed) → feedback loop back to Node 1.
Corridor types
- Discretisation — Continuous → categorical. Primary danger zone.
- Routing — Category → operational path.
- Feedback — Downstream outcome → upstream calibration. Enables drift persistence.
Resolving the AI Impedance Mismatch
The fundamental driver of agentic AI integration complexity is an architectural impedance mismatch. Legacy databases and rules engines rely on deterministic, spec-closed interfaces. Meanwhile, Generative AI systems produce probabilistic, semantically open outputs.
The ISE framework formalises this mismatch at the corridor level, proving that integration failures (often resulting in downstream structural and format failures) occur precisely where semantic latitude is violently quantised by a legacy threshold.
Core Constructs
The ISE Graph
Nodes: Systems, each a black box with defined input/output interfaces.
Edges (Corridors): Connections characterised by transformation operators capturing interface mechanisms: schema mapping, thresholding, formatting, truncation, routing logic.
Spec-Closed vs Semantically Open Interfaces
- Spec-Closed. Exactly one valid output Y(x) per input x. Classical deterministic systems. For every input x, |Y(x)| = 1. (Ayada, 2026, Def. 3.1)
- Semantically Open. Multiple valid outputs per input. AI systems, especially NL or probabilistic, are the primary source. ∃ input x where |Y(x)| > 1. (Ayada, 2026, Def. 3.2)
Critical Risk Clusters (CRC) and Blast Radius
Critical Risk Cluster: A connected subgraph with (1) at least one AI-source node with a semantically open interface and (2) coherence-risk score exceeding threshold for sustained duration (Ayada, 2026, Def. 4.2). Identifies where ambiguity can and is accumulating.
Blast Radius: The h-hop neighbourhood of a CRC: the set of downstream systems affected if propagation continues (Ayada, 2026, Def. 4.3). Provides a control variable for containment scope.
What Existing Tools Miss
| Approach | Monitors | Gap |
|---|
| MLOps | Model performance | Cannot detect post-boundary effects |
| Data quality | Input distributions | Drift occurs with stable inputs |
| Observability | Latency, errors | No semantic boundary awareness |
| ISE framework | Boundary telemetry | Purpose-built for inter-system drift |