Value 7 · founding principle
Human Authorship — authorship cannot be delegated
The seventh value of the AIAD framework is also its most structuring principle: authorship of intent is never delegated to the agent. The agent may enrich, question, refine — it may not initiate. Every Intent Statement must be written in the first person by an identifiable human.
What the value actually says
Human Authorship draws the boundary that makes AI-agent orchestration viable over the long term: the agent executes with excellence, but the intent always belongs to the human. This boundary is not a philosophical stance — it is operationalised by a mechanical rule: when in doubt about the intent, the agent ASKS, it does not invent. A guessed intent is treated as a betrayed intent. This is value 7 translated into the daily life of every session: an agent that "fills in" a missing WHY to keep the task moving commits the gravest mistake in the framework, even if the result looks plausible.
When the agent cannot decide, it emits a JNSP verdict ("I don't know") — a valid signal, never a failure.
The Human author field — the friction that marks ownership
The Human author field of the Intent Statement is mandatory and filled in deliberately — never automatically. If an agent wrote the Intent Statement, the session is invalidated. The Execution Gate cannot be passed without an Intent Statement whose author is human and identifiable.
This tiny friction is intentional: it marks the moment of ownership. Asking the agent to "draft an Intent Statement to save time" is the canonical anti-pattern — the time "saved" is paid back in intentional debt, then in drift. The framework's success indicators require an identifiable human author on 100% of Intent Statements.
Serious agentic programming vs vibe coding
External legitimation arrived in May 2026: the Human Authorship value corresponds precisely to what Martin Fowler calls "serious agentic programming" — the developer reviews and validates the generated code — as opposed to "vibe coding" (Karpathy/Fowler), where the author delegates without validation. It also prevents cognitive debt (Fowler/Joshi, 2026): the accumulation of code whose implementation is accepted without understanding the underlying conceptual model. What was not understood once will repeat itself — exactly what the Human Learnings section of the AGENT-GUIDE documents.
The field evidence: 29,585 PRs
The principle is not only defensible in theory — it describes what the industry already does at scale. Across 29,585 pull requests involving agents, merge authority remains "almost exclusively human". The human keeps the last word on what enters the product, even when the agent produced most of the code. AIAD formalises this checkpoint rather than leaving it implicit: the Execution Gate on the way in, human validation and merge on the way out.
The authority continuum
Human Authorship is not limited to writing the Intent Statement. Since doctrine v1.9, the value unfolds into an authority continuum: each move from one level to the next is a distinct human act that no other act replaces. This rule applies value 7 without changing its scope.
| Level | Required human act | What it does not authorise |
|---|---|---|
| Consultation | None — the agent may read and propose | Any writing of code or production artifacts |
| Clarification | Answering the agent's questions | Answering is not consenting: no writing of code or production artifacts follows from it |
| SPEC validation | Opening of the Execution Gate (SQS ≥ 4/5, Stranger Test) | Deployment |
| Write authorisation | Launching execution on the validated SPEC | Any action outside the SPEC's scope |
| Deployment authorisation | Explicit decision, complete chain of evidence | Undeclared irreversible actions |
"Writing" here means code and production artifacts. At the consultation and clarification levels, the agent may draft specification artifacts — that is the role of /sdd spec; they commit nothing until a human has validated them at the Execution Gate. The Intent Statement, for its part, is still written by a human.
Stop points at the agent's initiative
The continuum describes what the human grants; we also need to describe when the agent must come back to the human. AIAD adopts the typology proposed by Ethan Mollick in 2026: the agent asks for approval (before an action outside its mandate or an irreversible one), an expert opinion (when the decision goes beyond the SPEC), a plurality of viewpoints (when several interpretations of the intent are plausible), or simply makes the human a witness to a notable decision. The SPEC may specify which ones apply, in its optional "Agent stop points" section. Greater autonomy is not a goal in itself: it is a variable you calibrate, SPEC by SPEC. The details — execution scope, stop threshold, isolation — are on the Agent autonomy page.
Accountability that can be demonstrated, in both directions
AIAD adopts the thesis put forward by Bruce Schneier and Nathan Sanders in June 2026: an organisation answers for the agents it deploys and should not be able to hide behind "the AI did it" — this is a thesis, not settled law. If the organisation answers for its agents, it must be able to show what it asked of them (the Intent Statement and its author), within which limits (the SPEC), who validated it (the Execution Gate trace) and what was delivered (Drift Lock).
These artifacts do not guarantee the outcome of a dispute. They document the organisation's diligence and make its accountability demonstrable — in both directions: a defective SPEC validated at the Gate commits whoever validated it. A trace that could only exonerate would not be evidence; that is precisely what makes it credible (values 2 and 6).
A direct answer to the AI Act (Article 14)
Human Authorship is a direct answer to the human-oversight requirements imposed by Regulation (EU) 2024/1689 — the EU AI Act — for high-risk systems (Article 14). Where the regulation requires that a human be able to oversee, understand and correct the system, AIAD structurally guarantees that the original intent is human, archived and verifiable. The AIAD-AI-ACT governance agent makes it a Tier 1 control point, with veto power.
The human-authorship-check skill
The principle is enforced mechanically by a dedicated skill, triggered by /sdd intent, /sdd init and /aiad intention. It checks five points before any strategic artifact is recorded:
| # | Check | Failure = signal |
|---|---|---|
| 1 | The Author field is an identifiable human name (not "Claude", "Agent", "AI", "TBD") | Non-human author |
| 2 | The 5 fields come from the user, not from a paraphrase by the agent | Agent-dictated intent |
| 3 | "WHY NOW" cites a concrete trigger (event, metric, deadline) | Immature intent |
| 4 | "OBJECTIVE" contains at least one measurable metric | Vague objective |
| 5 | "DRIFT CRITERION" is an observable signal, not a rephrased intent | Inoperative safeguard |
Three verdicts are possible: PASS, FAIL (a known error with an actionable correction — e.g. author "Claude" → ask for the human name), or JNSP (the human is stuck on "WHY NOW": the intent is not ripe, the writing is postponed). A JNSP is here a success of the skill — it has detected an area that must remain human.
The Intent Workshop — a purely human space
The ritual counterpart of this value is the monthly Intent Workshop: 60 minutes, purely human, no AI. The /aiad intention command only serves to prepare it (compiling the month's Intents, alignment metrics, Human Learnings) — never to run it. Even in --fast mode, the agent only skips the explanations about the ritual, never the human facilitation itself. The workshop's question sums up the whole value: "are we still building what we wanted?"