Your agents do the work. A person answers for each one.
EDAVEN Organization & Agents is the operating system for your organization of people and agents: the four branches the market recognizes, brought into operation with receipts, plus a fifth, accountability.
- QueryReads and answers. Changes nothing.
- Reviewable writeExecutes, and a person reviews afterwards.
- Proposal with approvalOnly happens once approved.
Platforms are ready for agents. The organization still has no method.
Every company is improvising how people and agents work together. Few turn it into value that shows up in results.
Snowflake positions itself as the “control plane for the agentic enterprise.”
Snowflake, press release, Apr 21, 202684% of companies haven’t redesigned jobs to fit AI.
Deloitte Insights, Apr 2, 2026McKinsey describes the “agentic organization”: a new operating model with people and agents side by side.
McKinsey & Company
The answer isn’t one more agent. It’s a sociotechnical organization running on a governed operating system.
Five branches, brought into operation.
Regulation, standards, roles and controls tell you what to govern. In EDAVEN, each branch becomes a screen, an act with an owner and a receipt. The fifth branch organizes who answers for every agent and every decision.
| Branch | What it is | Why it matters | In EDAVEN |
|---|---|---|---|
| Regulation | The laws that apply to your use of AI. | Fines, suspension and legal liability start here. | EU AI Act tied to the controls that meet it. GDPR and LGPD tied to data and access. |
| Standards | AI management and security standards. | They are the yardstick auditors and clients will apply. | ISO/IEC 42001 and NIST AI RMF. OWASP for LLMs and agents, MITRE ATLAS. |
| Roles | Who decides, who approves and who answers. | Without a named owner, nobody answers when the agent fails. | Every agent has an owner. Role-based access (RBAC). Governance groups. More access needs dual approval. |
| Controls | What is applied to every act of the agent. | It is where policy becomes practice, or stays on paper. | Inventory of agents, models and third-party AI. Dual approval for high risk. Evaluation before release: adversarial and bias testing. Actions outside the mandate denied before they run. Audit, provenance, stops and spend limits. |
| AccountabilityEDAVEN only | Who answers for each agent, by contract. | Agents act on your behalf; someone has to answer for them. | Delegation mandate accepted by the accountable person. Shared accountability when another team uses the agent. Progressive autonomy, based on evidence. Your agents in Claude, Codex and the IDE (MCP). Agents from other organizations (A2A). |
Reference mapping. Not legal advice or certification.
Your agents’ intelligence is yours. Judgment stays with people.
The market is starting to call this cognitive sovereignty: control over the memory, provenance and delegated authority that accumulate as your organization works with AI. In EDAVEN Organization & Agents, it is native.
The learning belongs to the organization
- Organization memory with human curation, author and history
- Knowledge and skills stay in your environment
- The learning stays outside the vendor’s model
Judgment belongs to people
- Final decision is always human
- Dual approval for high risk
- Progressive autonomy, based on evidence
“Its data may be sovereign. Its learning may not be.”
Bramley Maetsa, ITWeb, Sep 28, 2026
Sources: Sondergaard Group, AI cognitive sovereignty; ITWeb (Bramley Maetsa, 2026), learning sovereignty; Forrester (2026), cognitive sovereignty; Fundação Dom Cabral, soberania cognitiva
Ten dimensions for designing controls.
- Inventory and risk
- Understand agents, models, purposes and exposure.
- Policy and compliance
- Translate applicable requirements into ownership and controls.
- Evaluation
- Define quality expectations, test scenarios and review criteria.
- Security
- Limit access, protect data and address misuse.
- Lifecycle
- Review versions, authorize changes and retire unsuitable capabilities.
- Data and privacy
- Establish provenance, quality, purpose and retention conditions.
- Identity and access
- Link each action to an owner and a mandate.
- Monitoring
- Track drift, consumption, actions and observed effects.
- Responsible AI
- Assess impact risks and fairness criteria appropriate to the case.
- Audit
- Organize records and evidence for review and accountability.
Plus one cross-cutting layer: cognitive sovereignty. It runs across all ten dimensions: the learning stays with the organization and judgment with people.
These dimensions guide governance design. Available controls and capabilities are verified for each delivery. Technical references do not imply certification or automatic compliance.
The factory’s 12 stations
A practical view of ownership, controls and records.
- AI model
- Select the model and record purpose, data destination and owner.
- Data product
- Define meaning, provenance, quality, access and owner.
- Memory
- Establish curation, change authorization and retention.
- Agent
- Attach identity, ownership and a mandate to its work.
- Supervision
- Separate proposal, review and approval according to risk.
- Connections
- Authorize tools and delegation within scope.
- Security
- Define blocking rules and information protection.
- Stopping
- Define when to stop and who may authorize resumption.
- Quality
- Test behavior and review changes before use.
- Records
- Track actions, owners and authorizations.
- Observation
- Review progress, consumption, exceptions and intervention needs.
- Outcome
- Compare observed and expected effects to guide the next cycle.
Frequently asked questions
What is the Governance Blueprint?
It is how EDAVEN organizes the governance of people and agents: regulation, standards, roles and controls, the branches the market recognizes, plus accountability. Each branch shows up in the product as a screen, an act with an owner and a receipt.
What is cognitive sovereignty?
It is the organization’s ability to keep control over what AI learns from it and over its own judgment. In EDAVEN, what agents learn stays in your organization’s memory, knowledge and skills, curated by people, and the final decision stays with people.
Does my company’s data train the model?
No. EDAVEN does not train models on your data. Agents become more capable through memory, knowledge and skills approved by people in your organization.
Can I use the agents in Claude, Codex or my IDE?
Yes. Agent teams are available through an MCP gateway with OAuth. Each person sees only the tools the mandate allows, and every call is recorded.
Is this certification or legal advice?
No. The mapping to laws and standards is a reference that shows which controls meet each requirement. It does not replace legal assessment or certification.
Which part of your operation needs to move forward?
Start with a recurring decision or an outcome that matters. Together we identify the data, people and first capability required.