Private Alpha · Human agency evidence
Don’t detect AI.
Document human agency.
What did the human actually contribute?
AgencyThread uses evidence from the work process to show how much substantive human intellectual agency the evidence supports and how the human and AI actually worked together.
A person can use AI extensively and still be the one who framed the problem, challenged weak ideas, made the important decisions, checked the result, and took responsibility for what moved forward. AgencyThread is built to make that contribution visible.
Private Alpha — currently inviting pilot participants
Aster Vale Coffee — U.S. Market Entry RecommendationSynthetic demo
Human Agency Index90 · High Human Agency
Evidence Confidence: Moderate
Low0–39Moderate40–59Substantial60–79High80–100*
Collaboration Profile Experimental
1Human-Led Creation
2AI-Augmented Human Work
3Co-Intelligence
4AI-Led Collaboration
5Delegated Generation
The manager framed the decision, redirected AI’s analysis, and originated the entry design the recommendation is built around — while AI substantively developed the models, scenarios, and draft.
The core model
Two questions. Two answers.
AgencyThread does not try to compress an entire human–AI workflow into one number. It gives you two complementary answers because they describe different things.
01Human Agency Index
How much substantive human intellectual agency does the evidence support?
The Human Agency Index looks at how the person framed and directed the work, evaluated AI contributions, developed or integrated ideas, verified the result, and retained meaningful responsibility for the outcome.
It is not a percentage of words typed by a human. It does not reward prompt volume, time spent, or manual effort. It is an evidence-based interpretation of the human’s substantive intellectual role.
Low0–39Moderate40–59Substantial60–79High80–100*
Collaboration Profile Experimental
What pattern of human–AI collaboration does the evidence suggest?
The Collaboration Profile describes how the work was produced: whether the human led the development with AI assistance, whether both parties materially shaped the work, or whether AI performed most of the substantive generation under human direction.
Two pieces of work can have similar HAI scores and still represent very different working relationships.
Collaboration Profile Experimental
1Human-Led Creation
2AI-Augmented Human Work
3Co-Intelligence
4AI-Led Collaboration
5Delegated Generation
Together, they tell a fuller storyNot just how much the human mattered, but how the human and AI worked together.
A different premise
Word origin is not intellectual origin
A document can contain substantial AI-generated language and still reflect meaningful human intellectual work.
A person may have developed the argument, chosen the evidence, rejected weak suggestions, restructured the analysis, checked the sources, and made the consequential decisions. A percentage describing who produced the sentences tells you something about language origin. It tells you much less about intellectual contribution.
AI involvement and human agency are not opposite ends of one scale. Both can be high in the same piece of work.
Human intellectual agency High
High Human Agency / Low AI ContributionThe human is highly consequential while AI contributes less substantively.
High Human Agency / High AI ContributionBoth human agency and AI substantive contribution can be high.
Low Human Agency / Low AI ContributionNeither dimension is strongly evidenced in the work record.
Low Human Agency / High AI ContributionAI contributes substantially while human agency is less evidenced.
AI substantive contribution Low → High
The process
How AgencyThread works
1Capture the work
AgencyThread evaluates evidence of how the work developed: AI conversations, drafts and revisions, notes, source checks, artifacts, and meaningful decisions.
A finished document alone usually cannot show that process. AgencyThread looks behind the artifact to the evidence of how it was made.
2Evaluate human agency
The evidence is assessed across six dimensions of substantive human participation. These cover how the work was framed and directed, how AI contributions were judged and developed, how the result was assured, and who ultimately understood and stood behind it.
Each dimension is considered separately. Strength in one area does not automatically erase limited evidence in another.
3Understand the collaboration
The report brings together a Human Agency Index, an Experimental Collaboration Profile, Evidence Confidence, and a plain-language explanation of what the evidence shows.
Where the record does not support a conclusion, AgencyThread should say so rather than fill in the gap.
1Evidence
2Agency Interpretation
3Human–AI Collaboration Report
ChatDraftHuman commentRevisionSource checkFinal decision
Judgment, not motion
What AgencyThread does not reward
AgencyThread does not treat activity as a proxy for thinking. It does not score how many words someone typed, how many prompts they sent, how often they edited a draft, how long they spent, whether they manually retyped AI output, or whether they disagreed with AI simply to appear more involved.
A person can appropriately delegate a routine task to AI and still exercise good judgment, even if that workflow does not require a high HAI score. Conversely, a person who spends hours retyping or lightly editing AI output does not demonstrate more agency merely because the process took longer.
AgencyThread looks for judgment — not motion.
Not agency signalsPrompt count · Keystrokes · Time · Retyping · Cosmetic edits
Agency signalsFraming · Judgment · Reframing · Evaluation · Verification · Accountability
Practical contexts
Built for the way people already work with AI
Education & AssessmentHAIProfileConfidence
Human framing and understandingIllustrative report structure Education & Assessment
AI-permitted assignments make “did the student use AI?” much less useful as the primary question. AgencyThread is designed to help instructors understand whether a student meaningfully framed the problem, evaluated AI contributions, developed the work, and understood what they submitted.
It is intended for AI-permitted learning and assessment contexts, not as a disciplinary accusation tool.
Professional Knowledge WorkHAIProfileConfidence
Judgment behind the deliverableIllustrative report structure Professional Knowledge Work
Consultants, analysts, researchers, strategists, and other knowledge workers increasingly use AI inside serious deliverables. AgencyThread can document the judgment behind that work: what the person framed, challenged, verified, changed, delegated, and ultimately stood behind.
That creates a more informative record than a generic AI disclosure or activity log.
Individuals & PortfoliosHAIProfileConfidence
A credible record of the processIllustrative report structure Individuals & Portfolios
Researchers, writers, students, job applicants, and independent professionals may want a credible record of how they worked with AI.
AgencyThread is designed to show not merely that AI was used, but how the person shaped the work through it.
Trust & method
Built to be trusted, not merely believed
AgencyThread only interprets what the evidence supports.
If the record is too thin to responsibly evaluate, the system should withhold the result rather than guess. Evidence Confidence is reported separately so readers can distinguish the Human Agency result from the strength of the underlying record.
AgencyThread is also designed to avoid surveillance-style proxies. It does not treat typing speed, screen activity, or time-on-task as proof of thinking.
Read our approach to Trust & Method →Evidence before inferenceMissing evidence ≠ missing cognitionNo activity-volume proxies
Private Alpha
Interested in trying AgencyThread on real work?
We are currently running a small private Alpha with selected pilot participants. Free to join, no credit card required.
