People describe what they do. An advisor connects the evidence. DutyGraph turns those accounts into a shared map of people, duties, tasks and handoffs—so the team can decide what to improve and where AI could help.
Today: an advisor-led pilot. Our north star is a living operating record that connects evidence, decisions, approved changes and measured outcomes. External agent delegation is a future step.
Start with discovery and reviewed evidence. Strategy builds on that foundation; bounded agents and external governance are the direction ahead. Select an area to see its purpose and current status.
Capture → review → map → investigate → agree changes → review again. People remain responsible for decisions.
Status guide: Built = implemented for the pilot, with real-client acceptance still open. Partial = some pieces exist. Modeled = demo structures and fixtures. Concept = a future direction. These labels do not certify customer outcomes or live integrations.
1. Discover the inside
Built
What it does
An advisor researches the company, agrees a scope with leadership, and invites participants through private links. People describe their work by voice or text. AI drafts task cards; participants check them before the advisor reviews the map.
Related tool categories
Interviews and SOP tools Capture accounts and procedures.Task and process mining Use activity and system records.
Examples of related tools, not a feature-equivalence ranking, partnership or integration claim. Logos belong to their respective owners.
How it fits
Start with the people doing the work, including the context that software logs may miss. Asynchronous responses reduce scheduling friction. Customer system connectors are not required to begin discovery; configured research and AI providers support the workflow.
What is real today
Research, kickoff, private responses, transcription, task cards and review are implemented in the advisor pilot. A complete audit with real providers and client acceptance remains to be validated.
2. Add the outside
Partial
What it does
The advisor reviews a sourced business brief and industry context alongside the internal work map. Research is an input to discovery and strategy, with sources and unknowns kept visible.
Related tool categories
Market research Business, customer and competitive context.
Examples of related tools, not a feature-equivalence ranking, partnership or integration claim. Logos belong to their respective owners.
How it fits
The direction is to connect relevant changes to the duties and stages they might affect, so teams can investigate their operational impact.
What is real today
Business research and industry mapping are implemented. Continuous external feeds, automatic impact mapping and notifications are future work.
3. Explore strategy
Built
What it does
Frameworks such as Business Model Canvas, Five Forces, Theory of Constraints and Balanced Scorecard help an advisor examine the same company from different angles. Findings can reference duties and tasks for further review.
Related tool categories
Strategy frameworks Structure questions and decisions.Planning tools Track objectives and execution.
Examples of related tools, not a feature-equivalence ranking, partnership or integration claim. Logos belong to their respective owners.
How it fits
Reuse company evidence across framework runs. Tracked dependencies flag analysis that becomes stale when upstream inputs change. People review the outputs and decide what to do next.
What is real today
Sixteen framework contracts, provider runs and dependency handling are implemented. Outputs depend on available evidence; missing financial or market inputs remain gaps. Measured business outcomes are not yet established.
4. Define bounded assistance
Modeled
What it does
The agent manifest model connects a proposed assistant to a task, the software it would need, a proposed permission scope and an accountable reviewer. It describes an intended delegation boundary.
Related tool categories
Agent builders Implement an assistant.Identity and access systems Manage and enforce permissions.
Examples of related tools, not a feature-equivalence ranking, partnership or integration claim. Logos belong to their respective owners.
How it fits
Begin with an explicit job and a responsible person. The design aims to keep delegated access within approved human authority; enforcing that boundary requires external integrations.
What is real today
Manifest models and fictional vendor-shaped fixtures demonstrate the concept. Production connectors, access provisioning and runtime enforcement are not built.
5. Keep authority traceable
Concept
What it does
The long-term goal is to trace an agent back to approved work and an accountable human, with a reviewable record of issuance, changes and revocation.
Related tool categories
Governance and observability Policy, inventories and activity records.
Examples of related tools, not a feature-equivalence ranking, partnership or integration claim. Logos belong to their respective owners.
How it fits
Keep understanding a task, owning it, approving a proposal and granting external authority as separate decisions. A task approval alone must never be treated as access permission.
What is real today
Versioned evidence, confirmation hashes and audit records exist today. The full external agent lifecycle, including issuance and revocation, remains future work.
DutyGraph complements tools for research, process analysis, strategy and access control. For source-linked vendor profiles, explore the AI governance directory.
What AI can look for once the work is mapped
The value is in the questions the map helps you answer.
Connected evidence helps an advisor compare accounts across teams, investigate gaps and prepare better questions. AI works with selected evidence; its findings need review.
Find the constraint
Where might work be waiting? The map helps an advisor investigate handoffs and propose a possible constraint with supporting evidence. A missing owner is a clue, not proof. The team confirms with measurement before calling it a bottleneck.
Today: cross-team analysis with structural checks, bounded AI hypotheses and exact citations. Theory of Constraints is one of the sixteen strategy views.
Surface what the team misses
Two teams may use the same word differently, describe overlapping tasks or leave a handoff unclear. Comparing their accounts can reveal questions that are hard to spot within one team. The advisor takes those questions back to the people involved.
Today: ownership gaps, conflicting accounts and missing handoffs are surfaced for advisor review. Findings are hypotheses until the people involved confirm them.
Connect change to work
The longer-term goal is to connect an external change to the duties and tasks it may affect, then help people review a response. The illustration shows that future direction. Continuous signals and agent notifications are not available today.
Direction: this depends on the outside feeds and the review loop above. No autonomous operating loop exists today.
Break an initiative into work
An initiative such as improving delivery reliability raises concrete questions: which stages does it touch, which duties matter, and what tasks might change? Framework findings help structure that discussion. AI-marked tasks in this illustration are possible assistance opportunities, not deployed agents.
Direction: framework outputs already point at duties and tasks. Proposed assignments are drafts for leadership, never automatic.
AI-generated findings are hypotheses with evidence links. The advisor and the team confirm them. The application does not claim a measured bottleneck because it found a missing owner.
The ladder: from the whole company to a single action
Follow one illustrative piece of work from business context to individual actions. The organization and terminology vary by company; workflows can cross teams and duties. This is an example, not a required hierarchy.
CompanyHow does this business make money?Sells industrial parts to contractors and plants.
Value chain phasesWhat are the phases that turn a need into revenue?Develop accounts, quote, confirm order, fulfil, collect, keep the customer.
Team and org chartWho works here and how are they organized?Twelve people across sales, purchasing, warehouse and finance.
RoleWhat is this person's job?Procurement manager.
ResponsibilityWhat must this role make sure happens?Supplier records are complete and approved.
DutyWhat does this person keep doing to meet it?Maintain supplier records.
TaskWhat exactly gets done, and what does it produce?Check an incoming supplier packet against the checklist.
Actions and workflowsHow is it done, step by step, and where does it go next?Open the packet, compare, flag missing items, send the draft to Finance.
Illustrative fictional distributor. These examples explain the concepts; they are not customer evidence or a claim that every rung is a separate product screen.
The advisor journey
How an advisor runs it
Start with one team and an agreed discovery question. Participants can respond asynchronously; the advisor guides scope, evidence review and the client readout. Timing and usefulness still need to be measured in real pilots.
01
A company asks for an audit
An interested party wants to understand its own operation before it commits to AI. The advisor qualifies the engagement.
02
Research
AI runs public research on what the company does, its market and its structure. The advisor reviews it.
03
Executive meeting
Where the company stands, where it wants to go, what the team and org look like. AI drafts the agenda. The advisor leads.
04
Personalized outreach
The advisor uses leadership context and the roster to prepare participant invitations. Each invited person receives a private response link.
05
Asynchronous collection
Each person answers on their own time, by voice or text: my tasks, my duties, how I do them, where data comes from, where I send it.
06
Task cards
AI drafts readable task cards from the responses. The participant checks and corrects them before submission. Confirmation records their understanding of the work.
07
Review and connect
The advisor compares accounts, spots gaps and conflicts, and connects the cards into the map.
08
Deliver the Duty Graph
The advisor prepares the Client brief and reviews the report for its audience. Agreed commitments and measurements can support ongoing review in the workspace.
Built around what lasts
A useful record as AI changes
DutyGraph’s lasting value should be the quality of the operating record: what people do, the evidence behind it and the decisions they have reviewed.
Traceable evidence
Evidence before inference
Keep source accounts, task versions and review decisions connected. A polished AI answer is useful only when people can check what supports it.
A lasting work record
Models support the work
Today’s AI integration uses OpenAI, with provider settings configured per account. Supporting another model requires integration and testing; the operating record is maintained separately from model output.
Human accountability
People own decisions
The aim is to help teams agree improvements and eventually delegate bounded tasks. Accountable people review changes, define authority and judge outcomes.
Understand the work. Improve the business.Build toward delegation with clear human authority.
