How structured is our intent?
The clearer our what and our why, the better AI builds the how. The structured intent maturity model places each of our systems on a scale from 0 to 4. Most organizations are at level 0 or 1 for most systems.
Implicit
Intent lives in people’s heads and in the code.
What it looks like
- Nobody can say for certain why a behavior exists.
- Knowledge leaves when people leave.
- AI is prompted ad hoc, and its output is judged by whether it seems to work.
- Nobody has written down which decisions software may make on its own.
The risk
Every change, migration or AI-generated feature risks breaking a promise nobody wrote down.
To reach the next level
Pick one critical system. Draw its boundary: what it covers, what software decides and what people decide. Then write down its goal, its key guarantees and who owns them.
Documented
Intent is written down, but as prose that drifts from the code.
What it looks like
- Requirements and specifications exist in documents, tickets or wikis.
- They are written once and rarely updated.
- Spec-driven AI tools may be in use, with specs in free text.
The risk
The documents and the system quietly disagree, and nothing detects it.
To reach the next level
Turn the most important statements into structured, testable clauses with examples.
Structured
Critical systems have structured intent that AI builds against.
What it looks like
- Each critical system has a written boundary: its scope, what software decides, what it only recommends, and what only people decide.
- Goals, guarantees, prohibitions, priorities and examples are separate, testable parts.
- AI-generated code is checked against the intent, not just reviewed by eye.
- Examples in the intent become automated tests.
The risk
Intent is precise, but it is not yet clear who decided each clause or how strongly it is enforced.
To reach the next level
Record the origin and owner of every clause, and the evidence for each implementation.
Governed
Every clause has an owner, an origin and recorded evidence.
What it looks like
- Each clause says who decided it and whether it is a promise, an accident or a bug.
- Every department with a stake is represented, and the priorities record how conflicts between them were settled.
- Changes to intent are versioned, with reasons.
- Each implementation reports which promises are proven, tested or only assumed.
The risk
Quality of the intent itself is not yet measured, and production behavior is not fed back.
To reach the next level
Measure how complete and precise each intent is, and compare live behavior against it.
Continuous
Intent is measured, checked continuously and portable across technologies.
What it looks like
- The quality of each intent is measured, for example by how many deliberately broken versions it rejects.
- Production behavior is continuously compared with the intent, and gaps trigger review.
- Moving to a new technology or vendor means generating a new implementation from the same intent.
The risk
The remaining risk is the hardest one: whether the intent says what people truly want. That stays a human judgment.
To reach the next level
Keep the judgment sharp: review validity with owners regularly, and invest in people who learn fast.
Assess a system
Pick one of our important systems and answer for it. Be strict: yes only if it is true today.
Is the system’s intent written down anywhere, such as in requirements, specifications or documentation?
Is its boundary written down: what it covers, what software decides, and what only people decide?
Are its guarantees and prohibitions written as separate, testable statements, with examples?
Is new or AI-generated code checked against that written intent?
Does each rule say who decided it, and which role or department has authority over it?
Are changes to the intent versioned, with the reason for each change?
For each promise, do you know whether it is proven, tested or only assumed?
Is the quality of the intent itself measured, for example by how many deliberately broken versions it rejects?
Is live behavior compared against the intent, with gaps triggering review?
0 of 9 answered. Your level appears when you’ve answered every question.
How to use it
- List your ten most important systems: the ones that handle money, safety, customer data, or that AI is already changing.
- Place each one on the scale, using the assessment above. Be honest: a system is at the highest level where every statement is true.
- Pick one system and move it up one level. Match the target to the risk: level 2 is a strong first target for important systems, and systems that handle money, safety or personal data deserve level 3.
- Review the placements every quarter, alongside other risk reviews.
For the business case, see structured intent for leaders. For the people who do this work, see hiring intent architects.
You may share and adapt this model with credit: Structured intent maturity model, jaynryan, structureintent.com (CC BY 4.0).