# Agentic AI

AI Glossary · Agents & Automation
Canonical URL: https://aisat.uk/glossary/agentic-ai
Author: Lucy Gilmore, Assistant General Counsel, Product, Privacy & AI at Sprinklr | Practical Privacy, Product & AI Governance & Defensible Risk Decisions (https://www.linkedin.com/in/lucy-gilmore/)
Breadcrumb: [Glossary](https://aisat.uk/glossary) > [Agents](https://aisat.uk/glossary?category=agents) > Agentic AI

## TL;DR

> **Agentic AI** refers to artificial intelligence systems that pursue complex goals with autonomy rather than merely responding to single prompts. By combining reasoning models, persistent memory, planning loops, and external tool execution, agentic systems break multi-step objectives into tasks, inspect intermediate outcomes, and adapt when errors occur.

## How Agentic AI works in practice

Traditional generative AI operates as a stateless text-in, text-out function: a user submits a prompt and receives a completion. Agentic AI wraps one or more foundation models inside a runtime control loop that perceives state, formulates a plan, invokes external tools, observes the resulting environment changes, and iterates until a termination criterion is met.

Under the hood, an agentic system relies on four architectural pillars: a reasoning core capable of task decomposition and self-correction; a tool layer connected via schemas or the Model Context Protocol; short-term working state paired with long-term episodic or vector memory; and deterministic guardrails that bound permissible actions. When a tool call returns an API error or an unexpected database row, the system reads the error trace and reformulates its next step rather than failing outright.

For example, a UK energy supplier uses agentic AI to handle disputed commercial meter billings. Instead of drafting a generic holding email, the system pulls half-hourly smart-meter telemetry, cross-checks tariff schedules in SAP, identifies a faulty multiplier after a meter exchange, calculates the credit adjustment, and queues the refund for a billing supervisor's one-click sign-off.

## Why it matters and trade-offs

Agentic AI shifts enterprise return on investment from individual productivity nudges to end-to-end process execution. While a chat assistant might save an analyst ten minutes drafting a summary, an agentic system can resolve complete back-office tickets, reconcile financial exceptions, or triaging security alerts overnight, compressing multi-day operational queues into minutes.

The commercial risk lies in compounding error rates, unpredictable token bills, and autonomous side effects. If a single reasoning step is 95 per cent reliable, a ten-step autonomous chain succeeds only 60 per cent of the time unless intermediate verifiers are built in. Organisations should avoid fully autonomous agentic AI for high-precision, deterministic flows where a conventional state machine or a structured prompt pipeline delivers lower latency, lower cost, and total auditability.

## Related AI concepts

- [AI Agent](https://aisat.uk/glossary/ai-agent): An AI Agent is a software system powered by a large language model that perceives environment context, maintains state, formulates multi-step plans and executes actions across external tools, APIs and databases to complete a delegated goal.
- [Agentic Workflow](https://aisat.uk/glossary/agentic-workflow): An Agentic Workflow is a multi-step operational or engineering process in which a language model dynamically plans steps, calls external tools, critiques its own output and branches based on intermediate results within an orchestrated pipeline.
- [Multi-Agent System](https://aisat.uk/glossary/multi-agent-system): A Multi-Agent System is an orchestration architecture in which multiple specialised AI agents with distinct system prompts, toolsets and context windows collaborate through structured handoffs or a supervisor to solve complex workflows.
- [Human-in-the-Loop (HITL)](https://aisat.uk/glossary/human-in-the-loop): Human-in-the-Loop (HITL) is an operational governance pattern in which an AI system or agent prepares a recommended decision, communication or tool call, but pauses execution until a qualified human reviewer approves or edits it.