IDD concept: Measure alignment
The IDD Maturity Model
The IDD Maturity Model describes four levels at which an organisation can expand AI autonomy while preserving clarity, accountability and control. It measures capability rather than adoption, and progression between levels is gated by measured alignment rather than by the number of tools or agents deployed.
Capability, not adoption
Many maturity models for AI count things: tools licensed, agents deployed, lines of code generated. Those numbers describe activity, and it is easy to mistake activity for alignment. The IDD Maturity Model asks a different question, which is how much autonomy the organisation can grant while still being confident that what gets built reflects what was intended. The answer comes from Intent Fidelity evidence, and it differs for different kinds of work.
The four levels
-
Level 1: Supervised Learning
Universal conservatism. Every Risk Dial is fully engaged and every AI-generated artefact is reviewed by a person. The purpose of this level is calibration: learning where AI operates reliably, and embedding fidelity measurement as a structural capability rather than an occasional exercise.
-
Level 2: Selective Delegation
Autonomy begins to differentiate. Low-risk categories of work move to spot-checking while higher-risk work stays under full review, and every act of delegation is gated on evidence. Standardised measurement becomes critical infrastructure at this level, because decisions now depend on comparing fidelity across teams and types of work.
-
Level 3: Scaling with Sustained Alignment
A structural shift. Engineering effort moves from inspecting output to articulating intent, and the organisation gains substantial efficiency while preserving human authority over high-risk decisions. For many enterprises this is the optimal steady state.
-
Level 4: Continuous Optimisation
Workflows themselves are reconsidered in light of what AI can now do, and specifications evolve to be clear to both people and machines. Not every organisation will pursue or need this level.
Plateau is not failure
The model is deliberately not a ladder that every organisation should climb to the top. For many enterprises, Level 3 represents a mature equilibrium: the work that can safely be delegated has been, the decisions that must remain human have stayed human, and the organisation is capturing real value without taking on unmeasured risk. Choosing to remain there is a legitimate outcome of the model rather than an incomplete use of it.
Anti-patterns and regression
- Perpetual pilots: experimentation that never builds the measurement needed to justify moving beyond Level 1.
- Premature acceleration: expanding autonomy faster than the fidelity evidence supports.
- Symbolic governance: controls that exist on paper, such as review steps that are routinely waved through, without shaping how work is actually done.
Progression is neither linear nor inevitable. When uncertainty rises, perhaps because of a new model, a new domain or a fall in measured fidelity, stepping back a level is disciplined risk management. The organisations that capture enduring value from AI are distinguished by their capacity to scale autonomy while preserving control over what matters.
Using the model
For CTOs and senior leaders, the model is primarily an assessment tool: a way of locating the organisation honestly, identifying the measurement capability it needs to build next, and setting expectations with boards and regulators about how quickly autonomy should expand. It also gives delivery teams a shared language for why some work is delegated and some is not, and it connects naturally to the Intent Hierarchy, which is how an organisation at Level 3 keeps shared intent coherent at scale.
Where it comes from
The Maturity Model was set out in The IDD Maturity Model, published on 26 February 2026, and forms Chapter 11 of the ebook.