Scroll to walk through the pipeline stage by stage. See what happens, which agents run it, and how the whole system connects.
Scroll to begin
Framework architecture
KKickoff
Prioritisation Matrix
AAssess
GGreenlight
EEngineer
NNurture
TTrack
The AGENT Pipeline
AI GovernanceBoundaries
AI AdoptionTrust
AGENTIC Vault
The system
One pipeline. Two streams. A living dashboard at the centre.
The AGENTIC Framework is an operating model for AI agent adoption. Not a one-time project, not a maturity assessment. A system that tells organisations what agents to build and when, then keeps them running.
Every part connects. Scroll to see how.
Entry point
Kickoff: Where should we look?
Two passes. First, a fast task list sweep to capture every workflow across the organisation. Then deep conversations with the people who do the work, letting them talk through what they actually do.
The Assessment Agent scores this data and populates the Prioritisation Matrix. Workflows that earned a deeper look move into the pipeline.
Observe the workflow as it actually runs. Not the documented version. The real one, with every shortcut, workaround, and judgment call that lives in people's heads. Fix what's broken before you formalise it.
The Workflow Extraction Prompt processes transcripts into structured data. The Specification Generator translates it into a machine-readable spec with success criteria.
Evaluate what AI can reliably do. Design the collaboration model: which steps are AI-run, which stay human-led. Score across three category groupings. Traffic-light decision: green means go, but people make the call.
The Greenlighting Agent reads specs, produces first-pass scores, and flags thin data. If the spec needs revision, the Design Loop sends it back to Assess.
The specification becomes real. Every workflow passes through a parallel-run phase where the agent shadows the human process. Trust is built through data, not promises.
Governance checkpoints are verified during parallel-run. Escalation paths are tested. The parallel-run itself is the adoption mechanism.
Monitor every live workflow. Every human override is a data point. Every correction is an instruction. The Assessor watches, the Builder acts, the Vault grows.
The Reality Loop is always running: overrides reveal gaps, the Assessor captures them, the Builder feeds corrections back into Assess or Greenlight. The system learns.
Watch the capability frontier. When new capability lands, workflows that scored too low get resurfaced. The Matrix reshuffles. "Not yet" becomes "focus here." Track also owns the decommission path.
Without Track, the pipeline runs once and stops. With it, the system compounds. Every cycle is faster than the last.
Governance and Adoption run alongside every stage.
AI Governance isn't bolted on after something goes wrong. It's built into every specification, every scoring decision, every deployment checkpoint. AI Adoption isn't a training session at the end. It starts with how you talk to people about their work in Kickoff.
Both streams are embedded, not sequential. That's what makes them work.
Foundation
The Matrix and the Vault make it compound.
The Prioritisation Matrix is a living dashboard, continuously rescored by agents. Every workflow has a traffic-light status. The AGENTIC Vault stores every spec, score, build file, and override log. Documentation as infrastructure.
When someone leaves, every agent has a source file, spec, and build record. Nothing becomes a black box. The system survives the people who built it.
Ready to go deeper?
Read the detailed articles, or see the complete framework with templates and worked examples.