Layer 01 · Strategic
Outcome and constraint definition.
- What pays. What's regulated. What's protected. What's measured.
Delivery system
The AI-Native Transformation Loop.
Five named stages. One iterative loop. Every Union Black engagement, Build or Transform, is a path through this system. The shape of the engagement determines how many stages we run, not whether we follow it.
01
Diagnose
Score AI readiness across the five dimensions. Map the workflows that matter. Identify the highest-leverage opportunity. Surface the gaps that block it.
Output · Diagnostic report & opportunity map
02
Architect
Design the target operating model. Pick the deployment pattern. Specify orchestration, agents, models, data, and governance. Make migration paths explicit from day one.
Output · Architecture spec & operating model
03
Deploy
Build and integrate the system. Wire in observability, audit, escalation, and fallback. Run the pilot with the team that will own it. Govern from the first request, not the last.
Output · Live system in production
04
Enable
Transfer capability. Runbooks, training, live coaching. The team operates the system before we step out. Internal AI Centre of Excellence stood up where the scope warrants it.
Output · Capability handover
05
Iterate
Continuously refine. Migrate models as the frontier evolves. Extend capability against the roadmap. Optional Continuous engagement; otherwise, your team owns the cycle.
Output · Quarterly uplift
Companion framework
The AI Readiness Framework.
Diagnose stage runs against this framework. Five dimensions, scored 0–5, with the gap analysis underneath. The score is a snapshot: the value is the path it forces you to commit to.
01
Strategic clarity
Is the AI mandate explicit, sponsored, and tied to an outcome that pays?
02
Operational maturity
Are workflows documented enough to redesign?
03
Data foundations
Is the data accessible, governed, and trustworthy?
04
Infrastructure capability
Can the architecture actually host this?
05
Cultural readiness
Is the organisation ready to operate alongside autonomous systems?
System layers
What 'AI-native architecture' actually means.
Layer 02 · Orchestration
The control plane that routes work to agents and models.
- Vendor-agnostic. Observable. Policy-bound. Replaceable.
Layer 03 · Agent
The specialised systems that perform the work.
- Bounded scope, transparent reasoning, escalation paths, audit trail.
Layer 04 · Model
The frontier models or local-inference engines that power agents.
- Pinned by capability and constraint, not by vendor preference.
Layer 05 · Infrastructure
Compute, networking, identity, secrets, and data layer.
- Sovereignty and residency are architectural decisions, not afterthoughts.
Layer 06 · Observability
Trace every request, decision, and output.
- Without observability, the system is unmanageable and undefendable.
Layer 07 · Governance
Policy enforcement, audit, model registry, escalation, kill-switch.
- The Governance & Security Agent has veto authority by design.
Transformation Roadmap
Augment → Automate → Autonomize.
Augment
People remain the operators. AI accelerates them. Output rises; the org chart doesn't change. This is the baseline most organisations should target first.
Automate
Bounded workflows run autonomously under policy. People shift to oversight, exception handling, and capability extension. Headcount stops scaling with output.
Autonomize
Whole functions operate AI-native. The operating model is redesigned around autonomous systems with human supervision as the design principle.
Most organisations move through these phases sequentially over 6–18 months. Skipping ahead burns capital. Going slowly burns competitive position. The Strategic Diagnostic identifies the right pace for yours.
The methodology isn't the differentiator. The discipline of following it is.Union Black · Operating Principle
Recommended next step
Run the methodology against your business.
Build Discovery applies stages 1–2 to one workflow. AI Readiness Assessment applies stage 1 across the function. Architecture Engagement applies stages 1–2 across the function. Pick the size of the question you want answered.