Agentic AI
Agentic AI is a general term for AI systems that interpret goals and autonomously repeat the cycle of planning, executing, and verifying actions without requiring step-by-step human instruction.
Asking a chatbot a question and receiving an answer — this was the core of AI utilization in 2023–2024. Agentic AI lies beyond that. Rather than answering questions, it completes tasks.
Karpathy coined this shift "Agentic Engineering" in early 2026. At its core is a workflow called the PEV loop — a cycle that AI itself runs through Plan → Execute → Verify. In coding, for example, an agent first analyzes the task and formulates an execution plan, writes the code, runs tests to verify the results, and makes corrections if problems arise. Humans provide approval and course corrections at checkpoints along the way.
Looking at real numbers: at Stripe, an internal agent called "Minions" is merging over 1,000 pull requests per week. Zapier reported that 89% of the entire organization has integrated AI into their work. This is no longer an experimental phase.
The infrastructure supporting Agentic AI is also being developed. MCP for tool connectivity, A2A for inter-agent communication, and Agent Skills as the unit of task execution capability — these are combining to enable the automation of complex workflows that were impossible with standalone AI models.
Related Terms
AI Agent
An AI agent is an AI system that autonomously formulates plans toward given goals and executes tasks by invoking external tools.
Agentic Flywheel
The Agentic Flywheel is a self-reinforcing growth cycle in which AI agents autonomously execute tasks, and the data and feedback generated from those results improve the agents' own accuracy and decision-making capabilities, enabling them to be entrusted with an ever-increasing range of tasks.
AI Employee (AI Employee)
An autonomous AI agent that takes on a specific business role and continuously performs tasks in the same manner as a human employee. It differs from conventional AI assistants in that it holds a defined scope of responsibility as a job function, rather than simply responding to one-off instructions.
