Agentic AI

Agentic AI is a type of artificial intelligence that independently performs tasks to accomplish a goal. Instead of simply responding to individual commands or prompts, an AI "agent" can determine the necessary steps, perform those actions, and make adjustments as needed.

Most agentic AI systems use one or more AI agents powered by large language models. An agent may break complex requests into several smaller tasks, then use external tools to complete them. For example, it might search the web, access a database, run a program, or communicate with another application through an API. The agent can use the output of each action to determine what to do next

Examples of agentic AI platforms include:

• OpenAI Codex
• Anthropic Claude Code
• Google Gemini Enterprise
• Microsoft Copilot Studio
• Salesforce Agentforce
• GitHub Copilot

Agentic AI vs Generative AI

While agentic AI can take actions to accomplish a goal, generative AI primarily creates content in response to user input. For example, a generative AI service like ChatGPT might answer a question, write an email, or generate an image from a description. Agentic AI goes a step further by taking actions on behalf of the user. Rather than only explaining how to complete a task, it may use available tools to complete the task itself.

For example, you might ask an AI chatbot to recommend a hotel for an upcoming trip. A generative AI service may provide a list of hotels based on your preferences. An agentic AI system could search current hotel availability, compare prices, check your calendar, and even make a reservation after receiving your approval.

NOTE: You can typically configure an AI agent with a specific level of autonomy. For instance, in the example above, you might allow the agent to perform all the research steps, but not make the reservation. Because agents can interact with external applications and make decisions without constant supervision, they often include safeguards that limit what information and tools they can access.