Structured intent for leaders
How structured is our intent? It is the question to ask about every system AI touches. The clearer our what and our why, the better AI builds the how.
AI has made building software cheap and fast. That is good news, with a catch: a wrong decision now gets built just as quickly and just as well as a right one. The scarce resource is no longer engineering time. It is clarity about what the organization actually wants its systems to do.
Structured intent is how an organization governs what AI builds. It is a written, owned record of what each important system is for, what it promises, what it must never do, and who decided. AI builds from it. Machines check against it. People remain accountable for it.
What it means in business terms
| The idea | What it means for you |
|---|---|
| Intent | What we decided the system should do, and why |
| Origin and authority | Accountability: who decided each rule, and when |
| Priorities | Our trade-off policy, such as risk versus speed or privacy versus convenience |
| Prohibitions | Hard limits and compliance boundaries that hold even under pressure |
| Evidence | Assurance: which promises are proven, which are tested, which are only assumed |
| Lifting | Recovering the business rules buried in legacy systems before they are replaced |
| Lowering | Freedom to change technology or vendor without losing what was built |
Why it matters now
- AI amplifies unclear decisions. When an AI agent misreads what you meant, it does not hesitate. It builds the misunderstanding at full speed, often in ways that pass every test.
- Accountability is becoming a requirement. Regulators, auditors and customers increasingly ask who decided what an automated system does and how the organization knows it works. A structured intent answers both questions in one place.
- Legacy modernization is a knowledge problem. The hardest part of replacing an old system is not writing new code. It is discovering which of the old system’s behaviors are promises the business relies on, and which are accidents. Lifting intent makes that discovery explicit and reviewable.
- Technology choices stop being one-way doors. When intent is written down independently of any implementation, moving to a new language, platform or vendor means generating a new implementation from the same intent, not starting over.
Boundaries come first
Before anyone writes down what a system should do, decide where its edges are. Security teams already do this: an authorization to operate starts by defining the system’s boundary, because nothing can be assessed until it is clear what is inside. Structured intent starts the same way:
- What it covers: what this system is responsible for, and what it explicitly is not.
- Who decides what: which decisions the software makes on its own, which it only recommends to a person, and which only a person may make. This is the question every organization adopting AI agents must answer, and few have written it down.
- Who rules: which role has the authority to decide each part of the intent.
- What it trusts: what the system relies on from outside without being able to check it.
The answer to “what are we letting AI decide?” should be written, owned and reviewable, not discovered after the fact.
Lines of sight
Every important decision should have a line of sight down to the code that carries it out, and every line of code a line of sight up to the decision it serves. Structured intent keeps those lines as part of the work rather than in a document beside it: each layer’s intent refines the one above it, from outcomes to systems to components to code.
- Goals flow down. An outcome the business commits to becomes promises in each system that delivers it.
- Evidence flows up. Ask whether “customers are never charged twice” is actually enforced, and the answer comes from the evidence at every layer beneath it.
- Change flows through. When a goal changes, impact analysis follows the lines of sight: which promises in which systems are affected, who owns them, and which evidence must be re-established. AI drafts the changes; the owners decide.
How big is this change? becomes a report, not a week of meetings.
It takes a team of humans, not just teams of agents
AI agents can and should draft intent: proposing clauses, suggesting boundaries, flagging contradictions. But every enterprise is unique, and only the people in its roles know its context. Structuring intent is a team effort in which each role judges the intent and represents its department’s interests:
| Role | What they bring |
|---|---|
| Intent owner | Business authority to rule on what is a promise, an accident or a bug |
| Intent architect | Structures the intent and brings the right people to each decision |
| Domain experts | Knowledge the code doesn’t contain, and their department’s interests |
| Engineers | Implementations, and evidence of how well each promise is kept |
| Assurance | Judgment on whether that evidence is strong enough for the risk |
| Governance and risk | The boundaries software must not cross, and the audit that it doesn’t |
When departments want different things, the intent’s priorities record how the enterprise chose, and who chose. Agents propose; people decide.
Six questions to ask your teams on Monday
- For our most critical systems, where is the intent written down, and who owns it?
- Which decisions are we letting software or AI make on its own, and where is that boundary written down?
- When AI generates code for us, what is it checked against?
- Which of this system’s promises are proven, which are tested, and which are just assumed?
- If we replaced this system next year, how would we know which of its behaviors we must keep?
- When two goals conflict, such as speed and safety, where is our trade-off decision recorded?
If the answers are “in people’s heads”, “nothing in particular” or “we’d have to ask the original developer”, there is work to do.
Where to start
- Find out where you stand. Assess your most important systems on the maturity model’s scale from 0 to 4. It takes two minutes per system.
- Draw its boundaries. Pick one system that matters: critical, already touched by AI, or due for modernization. Agree what it covers, what software decides and what people decide.
- Assemble the team. Name an intent owner with the authority to rule, and bring in each department with a stake. Then lift the system’s intent before anything else changes.
- Build the capability. The people who do this work well are a distinct profile. See the workforce framework for the roles and competencies, and hiring intent architects for what to look for.
The manifesto sets out the full philosophy. The code repository shows a complete worked example for technical teams.
AI builds the software quickly. Structured intent makes it the right software.