Questions about their work.
Each person receives a private link and prompts based on their role and the leadership conversation.
The connected company workspace
Advisor pilotMap how work happens before you delegate it to AI. An advisor connects people, duties, tasks, and software—starting with one team inside your business.
Sign up for a demo Walk through a team responseSeeking 5–10 pilot companies.
Help us test discovery with a real team.
A clearer view of the business
Follow a piece of work. See the people behind it. Open the details that explain what happens next.
Screens show illustrative work. AI labels describe proposed roles; software labels describe the tools used in a task.
Discovery that fits around the work
Hold the leadership kickoff together. Then let each person explain their work when they have time. Keep live conversations for the questions that need them.
Each person receives a private link and prompts based on their role and the leadership conversation.
Voice is preferred: a real example brings out steps, exceptions, and frustrations. Typing works too. Check the task cards made from your answer before sending.
The advisor receives each person’s answer and reviewed task cards together, then checks gaps and differences across the team.
We are testing completion times and the quality of the resulting work map with our first pilot teams. Timing depends on team participation and follow-up.
Try the five-person walkthrough ↗From scattered knowledge to shared understanding
The useful details live between people.
Who checks the work. What the next team needs. Where a decision waits. What everyone assumes somebody else owns.
DutyGraph brings those connections into view, so the next improvement starts with how the business actually works.
From the first conversation to a clear work recordThe work record
Give people and proposed AI agents the same clear starting point: the job, the expected result, and the boundaries.
Inputs, instructions, and an expected work product give the next person something clear to receive.
Distinguish human work, proposed AI work, and AI work that requires human review. Make the checkpoint explicit.
See supporting evidence and version history. When the work changes, review the description with the people who do it.
How it works
One company. One engagement.
A clearer record at every step.
Collect company, market, competitor, and customer-community sources. Prepare for the first conversation with useful context.
Public research → Meeting preparationCapture goals, departments, and responsibilities. Turn the kickoff into a proposed team roster you can review.
Leadership input → People & dutiesSend personal questions through private links. Each person describes their work and checks the task cards made from their answer.
Team responses → Participant-reviewed descriptionsReview task descriptions with their people. Explore handoffs, test possible constraints, and define where AI could help.
Confirmed work → A plan to improveFrom understanding to action
Discovery is the beginning. Use the record to ask better strategic questions and review what changes.
Connect business models, market forces, customer needs, and possible constraints. Ask the strategic question, then inspect the evidence behind the answer.
Review open questions, decisions, and progress. Keep measurements and follow-up work connected to the issue they are meant to address.
Prepare readable reports with findings, sources, and next steps. Keep a dated snapshot of what was approved as the company record develops.
A few useful answers
DutyGraph starts with an advisor-led discovery engagement. The advisor helps your team understand and improve its work. A trained internal owner can later maintain the record, with ongoing advisor support as needed.
Yes. Create an account and open your private fictional Cobalt sample company. It includes people, duties, task cards, and example workflows. Nothing is sent to anyone when you open the sample.
The current pilot helps you define and review proposed AI work, tools, and human checkpoints. It does not run agents inside your business systems. A software logo on a task describes the tool used; it does not mean that tool is connected.
Add your provider keys in Workspace settings: Exa for research, OpenAI for AI drafts, and Resend for invitation emails. Email also needs a verified sender. Provider usage is billed by those providers. You can explore the sample and maintain records before connecting them.
AI output is a draft. Sources remain attached to the work, uncertain claims remain visible, and people review and confirm task versions. A proposed duty or a possible constraint is not silently treated as an established fact.
A branded email with their questions and a private response link. On their first visit they set up protected access. Voice is preferred, and typing is available. They check task cards created from their answer, make corrections, and send both together. The advisor then reviews gaps and differences across the team.
THE GOVERNANCE FIELD GUIDE
Plain-language guides to AI governance, agent boundaries, and the tools that do different parts of the job.
Connect governance decisions to real tasks and the people responsible for them.
Read the guide →A source-linked map of governance and identity platforms, with DutyGraph in context.
Explore the map →An editable manifest worksheet for the owner, task, scopes, and review conditions.
Get the worksheet →DutyGraph + Signet / Human authority
Identity tells you who. Access tells you what. DutyGraph connects the work to why an agent needs access—and on whose authority.
Map each person's duties and tasks. Record the tools the work touches and the actions it needs.
A manifest proposes a limited role for an agent. A designated reviewer checks the exact work and scope before authorization.
Carry the owner, task versions, requested access and review history in one record. Proposed scope stays within the person's permitted delegation.
Explore sample Workday, Oracle HCM, Okta, Entra ID and Saviynt records. See how identity, employment and company policy inform the request.
Available now: requests, versioned manifest drafts and a connected sample. Live IAM provisioning, signed attestations and automatic revocation are integration work, not active features of this pilot.
START WITH A FOCUSED ENGAGEMENT
Agree one workflow, its people, and the evidence needed to make a useful decision.
Explore enterprise discovery →Learn the discovery method, practice in the sample, and prepare to lead a supervised pilot.
Explore the advisor program →We are shaping the delivery, engineering, and security capabilities this work needs.
See where you could contribute →Founding pilot / 5–10 companies
Sign up for a demo of advisor-led discovery. If it fits, we will agree the scope, schedule, and any fees before the pilot begins.
We are recruiting our first pilot companies. Results and turnaround times are still being tested.
A better place to begin
Meet the advisor. See the flow. Decide whether a pilot fits.