Agentic AI is moving rapidly from experimentation into operational business workflows. Futurum Group\u2019s AI Platforms Decision Maker Survey, which included 820 respondents in the first half of 2026, found that 72 per cent of organisations were piloting or deploying agentic AI. It also found that 55 per cent\u00a0identified\u00a0agent reliability and hallucination management as a leading adoption challenge, while 53 per cent pointed to data privacy and\u00a0security.\u00b9\u00a0<\/p>\n\n\n\n
The enthusiasm is real. So is the hesitation. <\/p>\n\n\n\n
In May 2026, Gartner predicted that by 2027, 40 per cent of enterprises would demote or decommission autonomous AI agents because governance gaps were discovered only after production incidents. Gartner warned that treating agent governance as a choice between complete restriction and complete trust creates avoidable failure.\u00b2\u00a0<\/p>\n\n\n\n
Governance done well does not simply constrain agentic AI. It creates the conditions for agents to\u00a0operate\u00a0safely, earn stakeholder confidence and scale sustainably.\u00a0<\/p>\n\n\n\n
OutSystems\u2019 2026 State of AI Development research, based on a global survey of approximately 1,900 IT leaders, found that 97 per cent were exploring agentic AI strategies. However, only 12 per cent had implemented a centralised platform for managing agent sprawl.\u00b3 <\/p>\n\n\n\n
This difference illustrates a broader operational problem. Many organisations are experimenting with agents faster than they are creating consistent controls for identity, access, monitoring,\u00a0testing\u00a0and accountability.\u00a0<\/p>\n\n\n\n
Projects can stall even when the underlying technology works. Business leaders, security teams, legal\u00a0advisers\u00a0and customers may be unwilling to rely on agents whose authority, behaviour and decision history cannot be clearly explained.\u00a0<\/p>\n\n\n\n
Human oversight should be based on the consequences of an action, not on a blanket rule applied to every agent. <\/p>\n\n\n\n
Stanford Digital Economy Lab\u2019s Enterprise AI Playbook found that the\u00a0appropriate level\u00a0of human oversight depends on error tolerance, regulatory\u00a0requirements\u00a0and task complexity. It also found that escalation-based models can work particularly well for high-volume, recoverable tasks, while approval models are better suited to regulated or high-stakes\u00a0work.\u2074\u00a0<\/p>\n\n\n\n
A practical governance model can classify agent actions into three tiers. <\/p>\n\n\n\n
Low-risk and reversible actions can run without a human approval step. Examples include retrieving approved information, drafting internal content, categorising requests and summarising service tickets. <\/p>\n\n\n\n
The agent should still operate within defined data, system and logging boundaries. <\/p>\n\n\n\n
Medium-risk actions should trigger a lightweight approval process. Examples might include sending an external communication, updating a customer record or changing a workflow status. <\/p>\n\n\n\n
The reviewer should receive enough context to understand what the agent proposes, why it proposes it and what systems or records will be affected. <\/p>\n\n\n\n
High-impact, regulated or difficult-to-reverse actions should pause until an authorised person approves them. <\/p>\n\n\n\n
Examples may include processing payments, changing access permissions, executing legal agreements, making employment decisions or modifying\u00a0sensitive production systems.\u00a0<\/p>\n\n\n\n
The agent\u2019s recommendation, evidence, approval decision and final action should all be recorded. <\/p>\n\n\n\n
There is no benefit in removing human involvement from a high-consequence decision simply to make a workflow appear frictionless.\u00a0<\/p>\n\n\n\n
One of the most important governance principles is separating an agent\u2019s technical capability from its authority to act. <\/p>\n\n\n\n
Agents should not automatically inherit the broad access of the employee, developer or team that created them. Instead, each agent should receive only the data, tools and actions\u00a0required\u00a0for its approved purpose.\u00a0<\/p>\n\n\n\n
A least-privilege approach should include: <\/p>\n\n\n\n
Microsoft\u2019s 2026 agent governance capabilities reflect this direction. Copilot Studio can provision Microsoft Entra Agent IDs for new agents, while Microsoft Agent 365 provides centralised visibility into agent inventory, permissions, behaviour and activity. Microsoft Entra Conditional Access can also apply access policies to agent identities before access tokens are issued. <\/p>\n\n\n\n
These capabilities can support governance, but technology alone does not determine which permissions are\u00a0appropriate. Organisations still need clearly defined owners, risk\u00a0classifications\u00a0and approval policies.\u00a0<\/p>\n\n\n\n
Governments and standards bodies are also focusing on agent identity, authorisation and traceability. <\/p>\n\n\n\n
Singapore\u2019s Infocomm Media Development Authority launched its Model AI Governance Framework for Agentic AI in January 2026 and updated it in May following feedback and contributions from more than 50 organisations. The framework recommends controls such as agent identity management, access controls, traceability and meaningful human accountability.5<\/sup> <\/p>\n\n\n\n In the United States, NIST launched its AI Agent Standards Initiative in February 2026 to encourage secure and interoperable agent standards. Related NIST work is examining how organisations can distinguish agent identities from human identities, limit agent authority, manage delegation and preserve accountability for automated actions.6<\/sup>\u00a0<\/p>\n\n\n\n The direction is clear. An enterprise agent should not be an anonymous process operating through an unmonitored shared account. It should have an identifiable purpose, defined authority, accountable\u00a0owner\u00a0and reviewable history.\u00a0<\/p>\n\n\n\n Technical controls are necessary, but they are not sufficient. <\/p>\n\n\n\n Resistance to agentic AI is often cultural and organisational rather than purely technical. Finance, legal, security and operational leaders may have previously seen automation programs create uncontrolled complexity, inaccurate\u00a0outputs\u00a0or unclear accountability.\u00a0<\/p>\n\n\n\n Their caution should be treated as useful risk information, not as an obstacle to innovation. <\/p>\n\n\n\n Four practices can help build confidence. <\/p>\n\n\n\n Allow stakeholders to observe an agent working with realistic data in a sandbox or read-only environment before it receives permission to take consequential actions. <\/p>\n\n\n\n This provides evidence of how the agent behaves without exposing the organisation to unnecessary operational risk.\u00a0<\/p>\n\n\n\n Choose a well-defined,\u00a0frequent\u00a0and recoverable task. Measure results such as time saved, accuracy, escalation rates, resolution times and human intervention.\u00a0<\/p>\n\n\n\n A measurable outcome is more persuasive than a broad claim that an agent has improved productivity. <\/p>\n\n\n\n Create a group that includes business, technology, security, privacy, legal, risk and operational stakeholders before a serious incident occurs. <\/p>\n\n\n\n Its role should include setting risk tiers, assigning ownership, approving high-impact use\u00a0cases\u00a0and reviewing performance and incidents.\u00a0<\/p>\n\n\n\n When a stakeholder asks what an agent did, the organisation should be able to provide a clear answer. <\/p>\n\n\n\n Logs should capture the agent\u2019s identity, instructions, data sources, tool calls, proposed actions, approvals, outputs and resulting system changes. <\/p>\n\n\n\n Audit information should be understandable to business and risk stakeholders, not only to developers.\u00a0<\/p>\n\n\n\n Governance cannot end when an agent enters production. <\/p>\n\n\n\n Agent behaviour can change as models, prompts, tools, permissions, data\u00a0sources\u00a0and connected systems are updated. A workflow that passed testing at launch may develop new risks after one of its dependencies changes.\u00a0<\/p>\n\n\n\n Production governance should therefore include: <\/p>\n\n\n\n Monitoring should examine both whether an agent completed its task and whether it completed the task within its authorised boundaries. <\/p>\n\n\n\n Gartner\u2019s forecast that 40 per cent of enterprises may demote or decommission autonomous agents is a warning, but it also points to an opportunity.\u00b2 <\/p>\n\n\n\n Organisations that\u00a0establish\u00a0proportional human checkpoints, task-scoped permissions, distinct agent identities, meaningful monitoring and transparent accountability will be better positioned to move successful agents into production.\u00a0<\/p>\n\n\n\n Those that treat governance as documentation to be added after deployment are more likely to discover their weaknesses through an operational, security or compliance incident.\u00a0<\/p>\n\n\n\n The\u00a0objective\u00a0is\u00a0not unrestricted\u00a0autonomy. It is controlled autonomy that can be observed, explained\u00a0and trusted.\u00a0<\/p>\n\n\n\n Insentra\u2019s\u00a0AI practice helps organisations design governance checkpoints, permission frameworks and stakeholder trust programs for enterprise AI. Explore Insentra\u2019s AI strategy resources through\u00a0AI Momentum<\/a>\u00a0to follow developments in agentic AI,\u00a0governance\u00a0and responsible adoption.\u00a0<\/p>\n\n\n\n Sources<\/strong> <\/p>\n\n\n\nBuild confidence with cautious stakeholders\u00a0<\/h2>\n\n\n\n
Start in a controlled environment <\/h2>\n\n\n\n
Make the first outcome measurable <\/h2>\n\n\n\n
Establish cross-functional governance early\u00a0<\/h2>\n\n\n\n
Maintain complete and usable audit trails\u00a0<\/h2>\n\n\n\n
Monitor agents after deployment\u00a0<\/h2>\n\n\n\n
\n
The stakes run both ways\u00a0<\/h2>\n\n\n\n
Ready to move from agentic AI enthusiasm to sustainable agentic AI governance?\u00a0<\/h2>\n\n\n\n
\n
\n
\n
\n
\n
\n