- Agentic AI systems plan, reason, and act toward goals independently.
- Enterprise adoption is accelerating, but over 40% of projects may fail.
- The term describes a spectrum of autonomy, not a binary category.
Agentic AI describes artificial intelligence systems that can pursue goals through independent planning, reasoning, and action, with minimal human oversight. Unlike conventional AI that responds to a single prompt, agentic systems perceive their environment, decide on a course of action, execute it, and evaluate the result before choosing what to do next.
Why It Matters
Key figure
$7.29 billion
Global agentic AI market value in 2025, projected to reach $139 billion by 2034
The distinction between agentic AI and earlier forms of artificial intelligence is not semantic. Traditional AI predicts. Generative AI creates. Agentic AI does both, then acts on the results. That difference changes what humans delegate to machines and what oversight those machines require.
The commercial stakes are large and growing. Gartner, the technology research firm, predicted in August 2025 that 40% of enterprise applications would embed task-specific AI agents by the end of 2026, up from less than 5% in 2025. The same firm projected that by 2028, AI agents would intermediate more than $15 trillion in business-to-business spending.
Yet deployment carries risk. Gartner also predicted in June 2025 that over 40% of agentic AI projects would be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls. Yoshua Bengio, the Turing Award-winning AI researcher at Mila in Montreal, has argued that AI agency, not raw intelligence, poses the real existential threat to safety.
How It Works
Agentic AI systems operate through continuous perception-reasoning-action loops. The system observes its environment (a database, a codebase, a conversation), reasons about the best next step, executes that step, then evaluates the result before deciding what to do next.
Key figure
1995
Year Russell and Norvig formally defined AI in terms of agents
The architecture builds on decades of work. Stuart Russell and Peter Norvig defined an AI agent in their 1995 textbook Artificial Intelligence: A Modern Approach as anything that perceives its environment through sensors and acts upon it through actuators. What changed in the 2020s was the arrival of large language models powerful enough to serve as the reasoning engine inside those agent architectures.
Modern agentic systems typically combine a foundation model (for reasoning and language) with tool access (APIs, databases, web browsers), persistent memory (to track context across sessions), and an orchestration layer that coordinates multiple specialized sub-agents. A coding agent, for instance, reads a task description, breaks it into subtasks, writes code, runs tests, reads error messages, and revises its approach without waiting for human input at each step.
The term "agentic" migrated into AI from psychology. Albert Bandura, the Stanford psychologist, used it to describe human self-directed behavior. Andrew Ng, founder of DeepLearning.AI, popularized the term in the AI context in early 2024. Ng proposed "agentic" as an adjective rather than classifying systems as agents or non-agents, arguing that agency exists on a spectrum rather than as a binary distinction.
Key Context
The governance gap. Singapore's Infocomm Media Development Authority released the world's first governance framework specifically for agentic AI in January 2026. The framework addresses a problem that existing regulations were not built for: the EU AI Act, negotiated before agentic systems proliferated, assumes AI that assists human decisions, not AI that makes and executes decisions independently. The EU's high-risk system rules take effect in August 2026.
The transparency deficit. The 2025 AI Agent Index, published by researchers documenting 30 deployed agent systems, found that 133 of 240 safety-related fields across browser and enterprise agents contained no public information. Only 4 of 30 agents provided agent-specific safety documentation.
FAQ
What is the difference between agentic AI and generative AI?
Generative AI produces content (text, images, code) in response to a prompt and stops. Agentic AI uses generative capabilities as one component within a larger system that plans, acts, evaluates results, and iterates toward a goal without waiting for new instructions at each step.
Can agentic AI systems make mistakes?
Yes, and the consequences differ from conventional AI errors because agentic systems act on their outputs. A 2025 METR study found that experienced developers using AI coding tools were 19% slower than those working without them, despite believing they were 20% faster. That perception gap illustrates how difficult it is to measure the actual impact of autonomous AI tools.
Who coined the term "agentic AI"?
The concept of AI agents dates to Stuart Russell and Peter Norvig's 1995 textbook. The adjective "agentic," borrowed from psychologist Albert Bandura's work on human agency, was popularized in the AI context by Andrew Ng in early 2024 to describe AI systems that act with varying degrees of autonomy.
Is agentic AI regulated?
Governance is emerging but uneven. Singapore published the first dedicated agentic AI governance framework in January 2026. The EU AI Act's high-risk provisions, taking effect in August 2026, will apply to some agentic systems, though the regulation was drafted before such systems became widespread.
Related Reading
Sources
- Primary Reference: Russell, S. & Norvig, P. (1995). Artificial Intelligence: A Modern Approach. Prentice Hall.
- Additional Context:
- Agentic AI, explained (MIT Sloan, 2025)
- Gartner: 40% of Enterprise Apps Will Feature AI Agents by 2026 (Gartner, August 2025)
- Gartner: Over 40% of Agentic AI Projects Canceled by 2027 (Gartner, June 2025)
- The 2025 AI Agent Index (arXiv, 2025)
- Singapore Launches Model AI Governance Framework for Agentic AI (IMDA, January 2026)
Fact Check: Claim-by-Claim Verification Verified
All 11 claims verified. Gartner predictions, Russell/Norvig definition, Ng popularization timeline, Singapore governance framework, and AI Agent Index statistics all confirmed against primary sources.
Sources used for verification
- Gartner: 40% of Enterprise Apps Will Feature AI Agents by 2026 - gartner.com
- The 2025 AI Agent Index - arxiv.org
- Singapore Model AI Governance Framework for Agentic AI - imda.gov.sg
- Agentic AI, explained - mitsloan.mit.edu
- Gartner: AI agents will command $15 trillion in B2B purchases by 2028 - digitalcommerce360.com
