World Economic Forum
AI Agents in Action
Clear agent guidelines, enforceable authorization profiles and a phased adoption life cycle from pilot to scale.
- Library section
- Companion references: Trust & governance
- Published
- May 2026
- Destination it supports
- An organisation able to delegate authority to AI agents "while remaining fully accountable for its actions", scaling from isolated pilots to governed portfolios of agents.
The playbook addresses “a decidedly practical question: how can an organization, in concrete terms, delegate authority to an automated system while remaining fully accountable for its actions?” Its foreword names the “gap between capability and authorization” as “the central challenge to large-scale adoption”. It provides “three core elements: clear agent guidelines, enforceable authorization profiles, and a phased adoption life cycle from pilot to scale”. Chapter 1 sets out enterprise agent guidelines: shared language for autonomy, authority and consequence; decision rights across the life cycle; when agentic systems are the appropriate solution; sequencing adoption; deployment contexts; and the human–agent operating model. Chapter 2 introduces the Agent Capability and Authorization Profile (ACAP), “a single, living document with seven sections” (identity and scope; operating context; authority and consequential events; controls and enforcement; evaluation evidence and promotion gates; monitoring, incidents and change log; sign-offs and re-authorization cadence), built across three phases: system design and assessment, prepare and deploy, and monitor and improve. An appendix gives an ACAP summary playbook listing actions, owners, completion criteria and required evidence. It is the third report in a WEF series on AI agents and is “written for practitioners who are piloting, evaluating and deploying AI agents”. Playbook PDF
Five Forces coverage map
How far the framework covers each force (coverage, 0 to 3) and how far it tells you what to do about it (actionability, 0 to 2).
-
Internal Pressures & Capabilities
Deep treatment
Implementable
5/5
Internal: coverage 3 of 3, actionability 2 of 2
Chapter 1 agent guidelines and checklist; 1.2 decision rights; 1.3 when agentic systems are appropriate; 1.4 sequencing adoption; 1.6 human–agent operating model (supervisory fatigue, automation bias, rotating approvers); ACAP seven-section structure (Table 1); ACAP execution framework with actions, owners, completion criteria and evidence (Table 2), from the Playbook PDF (Tier 1)
-
Global & Geopolitical Dynamics
Acknowledged
Directional
2/5
Geopolitics: coverage 1 of 3, actionability 1 of 2
Jurisdictions within the operating context; "agent passport" for global or cross-border use cases, presenting machine-readable claims of identity, authorisation and jurisdiction-specific compliance, from the Playbook PDF (Tier 1)
-
Societal & Regulatory Pressures
Deep treatment
Implementable
5/5
Society: coverage 3 of 3, actionability 2 of 2
Regulation built into the authorisation method: agents "lack clear legal status, moral responsibility and reputational incentive", so governance is built to offset these gaps through named owners; ACAP's modular structure "supports regulatory alignment", providing "documented evidence of compliance" with human oversight under Article 14 of the EU AI Act; alignment with ISO 42001 and the NIST AI RMF; risk tiers mapped to the enterprise risk taxonomy; "regulatory changes that redefine compliance" named as a trigger for updating the model, from the Playbook PDF (Tier 1)
-
Industry & Competitive Pressures
Acknowledged
Directional
2/5
Industry: coverage 1 of 3, actionability 1 of 2
1.5 deployment-context tiers: Level 2 multi-organisation single-platform ecosystems governed by shared platform rules; Level 3 multi-platform, cross-boundary interaction; buyer and vendor agents acting for distinct organisations under differing authority limits, from the Playbook PDF (Tier 1)
-
Human Needs & Aspirations
Acknowledged
No practical guidance
1/5
Human needs: coverage 1 of 3, actionability 0 of 2
Customer-facing interactions named among consequential events needing control gates; preserving "stakeholder trust" as an outcome of governed adoption, from the Playbook PDF (Tier 1)
Total 15/25
For a companion reference the total reflects Five Forces coverage, not quality within its discipline.
Strategic Center Evaluation
| Strategic Parameter | Framework’s Treatment | Rating |
|---|---|---|
| Time Horizon | Sets out “a phased adoption life cycle from pilot to scale” and guidance that early use cases be “deployable in weeks”, with no declared planning windows for the strategy (Playbook PDF). | Implicit |
| Automation vs Augmentation | Section 1.3 asks whether deploying an agent is “the right approach to begin with” against “simpler, more predictable solutions”, and Section 1.6 designs supervision to preserve the quality of human judgement. The choice is framed as agentic against deterministic automation rather than as automation against augmentation (Playbook PDF). | Implicit |
| Idea-Execution Dynamic | Recognises “bottom-up innovations” as employees build agents on low-code platforms and sets signals for bringing them into governed deployment, without naming the rising relative value of ideas (Playbook PDF). | Implicit |
Adjacent Publisher Context
The following materials are published by the World Economic Forum and address topics relevant to the Five Forces, but are not explicitly positioned as components of AI Agents in Action: A Playbook for Trusted Adoption, Authorization and Scaling. They are not included in the /25 coverage score above.
| Force | Adjacent Source (URL + date) | Relevance Summary |
|---|---|---|
| Internal Pressures & Capabilities | Navigating the AI Frontier: A Primer on the Evolution and Impact of AI Agents (Tier 2, 16 Dec 2024) | First report in the series; established shared definitions and the evolution of types of agents. |
| Societal & Regulatory Pressures | AI Agents in Action: Foundations for Evaluation and Governance (Tier 2, 27 Nov 2025) | Second report in the series; introduced agent classification, evaluation, risk assessment and proportionate governance. |
| Societal & Regulatory Pressures | Advancing Responsible AI Innovation: A Playbook (Tier 2, 2 Sep 2025) | Nine plays for organisation and government leaders on putting responsible AI into practice; assessed separately as a companion reference. |
| Global & Geopolitical Dynamics | AI Infrastructure in the Age of Sovereignty: Requirements, Strategies and a Trusted Framework for Digital Embassies (Tier 2, 2026) | Addresses sovereignty requirements and strategies for AI infrastructure, including digital embassies. |
| Societal & Regulatory Pressures | Making Agentic AI Work for Government: A Readiness Framework (Tier 2, 2026) | A readiness framework for agentic AI in the public sector. |
Narrative Strengths & Gaps
Internal Pressures & Capabilities. Internal capability is the playbook’s centre of gravity. The agent guidelines checklist sets shared terms for autonomy, authority, consequential events, operational context and boundaries; allocates decision rights so every deployment has named owners; and tests whether an agent is the right solution at all. The human–agent operating model addresses supervisory fatigue and automation bias with rotating approvers, attention-management protocols and escalation time limits matched to agent speed. The ACAP execution framework turns all of this into a phase-gated record of actions, owners, completion criteria and evidence. The ACAP can also serve as the basis for an agent registry, a record of the agent estate rather than of the organisation’s AI strategy as a whole (Map: Not met). The three life cycle phases give each agent deployment a governed route from design to monitored operation, but not a route for the organisation’s strategy (Course: Partly met).
Global & Geopolitical Dynamics. Jurisdictions form part of each agent’s operating context, and the playbook proposes an “agent passport” so that agents in global or cross-border use cases can present verifiable, machine-readable claims of identity, authorisation and jurisdiction-specific compliance. National AI strategies, sovereignty and geopolitical competition are outside its deployment-level scope; the Forum covers sovereignty in separate work (Section 2.75).
Societal & Regulatory Pressures. Regulation is built into the authorisation method rather than handled as a separate compliance exercise. Because agents lack legal status, moral responsibility and reputational incentive, the playbook supplies accountability through named human owners, defined authority grants and auditable evidence. The ACAP’s modular structure provides documented evidence of compliance with requirements such as human oversight under Article 14 of the EU AI Act, and the model is designed to operate within ISO 42001 and the NIST AI RMF, with agent risk tiers mapped to the organisation’s existing risk taxonomy. Regulatory changes that redefine compliance are named as a trigger for updating the model, and changes to scope, authority or risk tier trigger re-authorisation. The boundary is societal expectations beyond compliance: what the public and affected stakeholders expect of agents once legal requirements are met is not analysed.
Industry & Competitive Pressures. The deployment-context tiers recognise that agents increasingly act inside shared platforms and across organisations, from single-organisation deployments, to multi-organisation ecosystems under shared platform rules, to multi-platform interaction with no central authority. The playbook illustrates the authorisation conflicts that arise when buyer and vendor agents for different organisations meet. Market competition, competitive positioning and industry adoption dynamics are outside its governance focus.
Human Needs & Aspirations. The playbook names customer-facing interactions among the consequential events that need explicit checkpoints, and frames governed adoption as a way to preserve stakeholder trust. External human expectations (customer attitudes to agents, public sentiment and talent-market demands) are not analysed in their own right, consistent with its role as an authorisation instrument.
What this framework does not give you. A map of where the organisation stands across its AI strategy; the ACAP records individual agent deployments. Without a clear picture of where it stands, an enterprise cannot tell which moves matter most. A route for the organisation as a whole; the life cycle phases govern individual deployments. Without a route, even a well-chosen destination stays an aspiration. A way to choose your own destination; it supports accountable delegation to agents at portfolio scale. An enterprise that has not stated its own ambition cannot check whether this destination is the right one for it. A view of what customers, prospective employees and the public expect of an enterprise’s agents. Those judgements shape trust and talent.
Synthesis. AI Agents in Action: A Playbook for Trusted Adoption, Authorization and Scaling governs how agents are authorised once the decision to use them has been made. It serves chief risk officers, AI governance leads, enterprise architects and business owners who are moving agents from pilots to production and need a repeatable way to authorise, supervise and re-authorise them. Its distinctive contribution is the ACAP: a single, auditable record linking what an agent can do with what it has been authorised to do, with regulatory alignment built in. Used alongside a strategy framework, it deepens the trust and governance discipline for agentic AI. The Five Forces lens adds a map and route for the strategy as a whole, the geopolitical and sovereignty context, competitive dynamics in agent ecosystems, societal expectations beyond compliance, declared planning horizons, the choice between automation and augmentation, and the rising value of ideas as execution becomes cheaper.
Sources
Tier 1 (scored), recency-gate pass unless noted:
- AI Agents in Action: A Playbook for Trusted Adoption, Authorization and Scaling – May 2026; canonical anchor; weforum.org blocks automated requests, so manually verified by the Sapience team on 31 October 2026 (report PDF retrieved from this page)
- AI Agents in Action: A Playbook for Trusted Adoption, Authorization and Scaling (PDF) – May 2026; full text; source of all Section 2 evidence; manually verified by the Sapience team on 31 October 2026
Tier 2 (reported in Section 2.75, not scored), each manually verified by the Sapience team on 31 October 2026 via the report PDF:
- Navigating the AI Frontier: A Primer on the Evolution and Impact of AI Agents – 16 Dec 2024; first report in the series
- Advancing Responsible AI Innovation: A Playbook – 2 Sep 2025
- AI Agents in Action: Foundations for Evaluation and Governance – 27 Nov 2025; second report in the series
- AI Infrastructure in the Age of Sovereignty: Requirements, Strategies and a Trusted Framework for Digital Embassies – 2026
- Making Agentic AI Work for Government: A Readiness Framework – 2026
Tier 3 (third-party corroboration, not scored): none used.
Excluded on recency or quality (not scored): none cited.
This library is our opinion, based on each organisation's publicly published framework as at the date on that assessment, linked from the assessment page. It is offered to help you see a framework's coverage and its open dimensions, not as a comparison, ranking, or endorsement. The organisations named have not reviewed or endorsed these assessments. Sapience provides advisory services that sit alongside, and sometimes in contrast to, the work assessed here.