Agentic AI definition
Agentic AI refers to AI systems that can pursue goals and take actions autonomously. Instead of only responding to prompts, they plan tasks, call tools and APIs, interact with apps, observe results, and iterate toward an objective, often with optional human oversight.
See also: training data, machine learning, end-to-end (E2E) learning, data augmentation, LLM temperature
How does agentic AI work?
- 1.Receives an objective.
- 2.Breaks the objective into steps and orders them.
- 3.Executes steps via function calling, APIs, or apps (email, browsers, code runners).
- 4.Stores intermediate results and updates plans as it learns.
- 5.Evaluates outcomes, reflects, and retries or adapts.
- 6.Seeks approval for sensitive actions when configured.
Agentic AI use cases
- Book a doctor’s appointment: Find nearby slots, fill out the intake form and schedule, and add them to your calendar.
- Return an online purchase: Locate the order, start the return, generate a label, schedule pickup, and track the refund.
- Monitor a product and buy on sale: Watch price drops, apply coupons, and place orders within a budget (with approval).
- Rebook a delayed flight: Detect the delay, find alternatives, message the airline, confirm the new ticket, and update your itinerary.
- Save energy: Learn your routine, adjust thermostat and lights, and report estimated savings.
