From visibility to initiative
Four stages. Standalone value at each. Architectural continuity across all.
Organisations do not need to commit to full governed intelligence to start. They need to see what they have. The journey begins with the MRI - making the invisible visible - and progresses only as far as the organisation chooses to go. Every stop along the way is a product in its own right.
The design principle
Build Stage 0 as if you might stop there. Architect it so that every subsequent stage is an extension, not a rebuild.
Some organisations will stay at the first stage and use the visibility for risk and planning. Others will use what the visibility reveals - complexity, risk, undocumented dependencies - as the business case for modernisation. A subset will carry the governed substrate forward into agent operations. The architecture must support all three paths from day one.
Stage 0 — MRI
Operational Visibility
Backward engineering. AI-assisted extraction of structured claims into a governed knowledge graph consumed by humans. No agents.
Stage 1
Accelerated Modernisation
Forward engineering. The governed graph guides the modernisation build. Read-only diagnostic becomes active reference.
Stage 2
Governed Intelligence
Agents become Players. Full governance apparatus comes online. Three-loop sustainability begins.
Stage 3
Agentic Operations
The shift from autonomy to initiative. Agents identify what to pursue, not just how to execute. Intelligence as emergent property.
Stage 0
MRI — Machine-Readable Intelligence
Operational visibility through extraction. Backward engineering of the systems already in production.
What it is
AI-assisted extraction of structured claims from existing systems - code, documents, procedures, operational records - into a governed knowledge graph consumed by humans through dashboards, reports, and structured queries. No agents. No autonomous decisions. All Players are human. The graph serves human understanding.
What it delivers
Total visibility into operational reality. The organisation sees its own structure - business rules, dependencies, exception patterns, regulatory constraints, technical debt - in a form that was never previously available. This is the MRI: it makes the organisation's intelligence machine-readable for the first time.
Every system integrator doing modernisation spends months on requirements gathering through interviews and document archaeology. The MRI compresses this by extracting what the system actually does and presenting it for practitioner validation. Practitioners correct and extend rather than recall from scratch. Faster, more accurate, and it catches the rules nobody remembers.
The validation loop
Even at Stage 0, a loop operates: extract claims from code and documents, present to practitioners, practitioners correct and extend, corrections feed back into the graph. This is the prototype of the engagement dynamic that later stages depend on - practitioners who see their corrections reflected in the graph will contribute when the stakes are higher.
Why this is a product, not a discovery phase
Traditional discovery completes, delivers a document, and ends. The MRI delivers a governed substrate that persists. The same graph that clarifies legacy reality at Stage 0 carries forward into modernisation at Stage 1 and into agent governance at Stage 2 - if and when the organisation chooses to continue.
Stage 1
Accelerated Modernisation
The governed substrate now informs the build. Read-only diagnostic becomes active reference.
What changes from Stage 0
The graph moves from read-only diagnostic to active reference. Developers and architects query it during the build. New claims enter as the modernisation uncovers additional business logic. The graph starts tracking both what is (legacy) and what should be (target), and the gap between them becomes a structured, queryable object rather than a spreadsheet of requirements.
Why the engagement loop holds
Practitioners are already engaged because Stage 0 showed them their operational reality reflected back. Stage 1 asks them to validate and extend what the MRI surfaced - and because they already trust the graph, the engagement loop gets its first real test under build-time pressure.
What it delivers
Faster, more accurate modernisation. Reduced requirements gathering overhead. A governed substrate that serves as living institutional reference during and after the transition. Better estimation - because the actual complexity is visible, not estimated. Fewer late-discovered business rules that force rework.
Stage 2
Governed Intelligence
Agents become Players. The full governance apparatus comes online. Three-loop sustainability begins.
The step function
Stages 0 and 1 serve only human Players. Stage 2 introduces agent Players alongside humans. The same architecture serves both, but agent Players trigger the full governance apparatus because their actions carry consequences at speed and scale that human review cannot absorb. This is the moment the operating model the venture is named for activates in full.
What comes online
Epistemic tiers, consequence classes, the circuit breaker, override governance with named accountable individuals, the agent identity model, the trace pipeline for epistemic feedback, escalation handling, and the practitioner visibility infrastructure that closes the engagement loop. The governance infrastructure transition is not gradual - you either have a circuit breaker or you do not.
The accountability gradient
The infrastructure is binary; the trust earned within it is progressive. Agent Players begin with Low-consequence authorisation - internal search, exploratory analysis - where the circuit breaker logs but does not halt. They progress to higher consequence classes only after the Domain Readiness Gate is passed and a named human reviewer is assigned. High and Critical consequence classes remain heavily human-supervised.
The compounding mechanism
Three loops begin closing in parallel. The economic loop reduces manual exception handling and error rates. The epistemic loop improves agent accuracy as traces deepen the substrate. The engagement loop reduces the cost of new domain knowledge as practitioners see their corrections reflected in production. Together they form the sustainability conditions for governed intelligence at scale.
Stage 3
Agentic Operations
From autonomy to initiative. The substrate compounds. The system is self-funding and self-improving.
From autonomy to initiative
Stage 2 agent Players are autonomous in how they execute within governed boundaries. Stage 3 agent Players begin identifying what to pursue - surfacing opportunities, anticipating exceptions before they occur, proactively challenging stale claims. The shift requires a deep substrate, mature dual-mode attention, rich trace reinforcement, and organisational trust earned through demonstrated governance quality.
Intelligence as emergent property
The system perceives institutionally relevant patterns, infers consequences, and surfaces action opportunities aligned with organisational purpose. New staff absorb institutional knowledge through the governed substrate rather than through years of apprenticeship. The system discovers patterns across domains that no individual practitioner could see.
The structural advantage
An organisation with a mature governed intelligence substrate operates in a fundamentally different mode from one running on documented procedures and institutional memory. Each additional domain is cheaper than the last. Cross-domain intelligence produces insights no single domain could generate.
Comparative view
Governance scales with consequence, not with ambition
Stage 0 needs basic epistemic governance: extraction confidence scoring, claim validation, practitioner review, version control. No agent Players means no action-consequence governance machinery. But the graph itself is crown-jewel information - business rules, dependencies, regulatory implementations - and asset-consequence governance must be proportional to that sensitivity from day one.
Stage 1 adds change governance as the graph evolves alongside the build. Claims about legacy and target states coexist; contradictions between them get typed and tracked. Still no circuit breaker, still no consequence classes.
Stage 2 activates the full action-consequence apparatus. Stage 3 adds initiative governance for agent Players that identify goals rather than execute assigned ones - an open area with limited precedent. The lesson clients should understand before committing: epistemic governance burden is roughly four times higher at Stage 2 than at Stage 0. Asset-consequence governance, by contrast, is proportional from Stage 0 onward and not deferrable.
Three paths the architecture supports
The same first stage, three legitimate destinations
The architecture does not assume the organisation will progress through all four stages. It assumes the first stage is valuable on its own merits and that any of three paths is legitimate.
Path 1
Stop at visibility
The MRI surfaces enough risk, enough hidden complexity, enough single points of failure that the organisation gets attributable value from mitigation alone. No modernisation, no agents.
Path 2
Modernise on the substrate
The visibility reveals the case for modernisation, and the governed graph becomes the spine of that programme. Faster, more accurate, fewer surprises during build.
Path 3
Carry it into agent operations
The substrate built and refined through Stages 0 and 1 becomes the governance foundation for agents at Stage 2 and the basis for initiative at Stage 3. The full journey.
Go deeper
The architecture and the research
The journey is the path through the architecture. The architecture page specifies the three primitives, the claim model, the circuit breaker mechanism, the four epistemic tiers, and the decay model. The research page links the published papers and manifestos that ground each part.