Blog / Governance

Agentic profiles: what the journal Nature proposes for governing agents

Rather than deciding whether a system is an agent or not, two researchers propose to describe it along 4 graded dimensions. The idea is useful; it leaves untouched the question of the power entrusted to the agent.

By Vincent Dorange, , 2 min read

4

dimensions to describe an agent: autonomy, efficacy, goal complexity, generality

On 12 August 2026 the journal Nature published online "Agentic profiles for effective AI governance", a perspective by Atoosa Kasirzadeh and Iason Gabriel 1. The paper starts from a simple observation: what is poorly described is poorly governed, and the binary label "agent" or "not an agent" does a poor job of describing the variety of systems in use.

Four graded dimensions

The authors propose to characterise an agent along 4 dimensions, each with its own gradations: autonomy, efficacy (its ability to act on its environment), goal complexity and generality 1. Each dimension raises its own questions about design, operation and governance. Combined, they produce "agentic profiles" ranging from narrow task-specific assistants to highly autonomous general-purpose systems; those profiles shed light on the technical and non-technical challenges of each class of agent.

The benefit is practical. Two systems sold under the same agent label may have very different profiles, so governing them the same way means watching one too closely or the other not closely enough.

What an agentic profile does not say

The profile describes what the agent is capable of. One question is left to each organisation: what decision power does it hand to the agent, decision by decision? Two companies can deploy the same agent, with an identical agentic profile, then give it powers of a completely different order. One lets it suggest discounts, the other lets it sign purchase orders.

The Agentic Commerce Framework® calls this the delegation profile, which is set out in the agent's mandate. The two readings complement each other: Nature says what the agent is capable of, the ACF says what it is allowed to decide. An organisation's agentic sovereignty depends less on how intelligent its agents are than on the power it delegates to them and on its ability to take it back.

Sources

Links to the official texts and original publications.

  1. 1Kasirzadeh A. and Gabriel I., "Agentic profiles for effective AI governance", Nature, vol. 656, pp. 320-328, Nature, published online 12 August 2026.

Glossary terms used in this article

AI agent mandateAgentic sovereigntyAI agent authority revocation

Vincent Dorange, creator of the Agentic Commerce Framework®

Also worth reading

95%

of organisations surveyed faced at least one agent-related incident in 12 months

Governance

95% of organisations have already had an AI agent incident, says Gartner

25

organisations interviewed; none lets its agents act with unrestricted autonomy

Governance

AI agents at work: what the OECD learned from 25 organisations

4 levers

transaction caps, budgets, approved counterparties and escalation rules, per Circle

Agentic economy

Agentic economy: agents that spend within delegated permissions

All articles