Agentic AI - what it is
Agentic systems engage, interact and influence rather than sit passively. AI agents focus on specific tasks with simple workflows; Agentic AI involves multiple agents running full end-to-end workflows with significant autonomy. They need infrastructure for autonomy, long-term memory and multi-step actions, plus risk models for emergent behaviours using frameworks like MAESTRO.
Agentic systems are active participants in digital environments → they engage, interact and influence rather than sit passively, demanding distinct infrastructure, risk models and governance.
| AI agents | Agentic AI |
|---|---|
| Focus on specific tasks with simple workflows → goal-driven, autonomous task performance | Involves multiple AI agents carrying out full end-to-end workflows with significant autonomy in more complex environments |
| Not new → think antivirus software and robotic vacuums | Newer → e.g., IT incident management, managing customer returns |
- Infrastructure must support autonomy, long-term memory and multi-step actions.
- Risk models → dynamic decision-making risk modelling with real-time monitoring, audit trails, explainability and human-in-the-loop/override mechanisms; must account for emergent behaviours via behavioural simulations, scenario-based modelling and multi-agent frameworks like MAESTRO.
- Governance must be dynamic, multi-layered and proactive → the three-tier guardrail framework.
99% of 1,000 surveyed enterprise AI developers were exploring or developing agents (IBM/Morning Consult) · Stanford's 2025 AI Index found agents already match human capability on select tasks with speed and cost advantages · and a major airline was held liable for its chatbot's misleading policy advice → the legal cost of insufficient guardrails.