Agentic AI is a developing technology that represents a shift from generative AI chatbots like ChatGPT.
Instead of relying fully on user prompts to take action (like generative AI does), agentic AI is autonomous, meaning it can act independently, take initiative, and perform tasks with minimal human input or intervention.
Agentic AI tools can help users with decision-making, problem-solving, and completing actions. They do this by acting in their digital environment, such as by filling out a form on a web page or editing a computer file. An agentic AI tool can improve its decision-making and performance over time by learning from its experiences.
At the moment, the most widely-available general agentic AI tools are internet browsers with agentic AI capabilities, such as Perplexity’s Comet. Some traditional browsers (like Google) are also beginning to incorporate elements of agentic AI.
GenAI generates new content based on user prompts, such as text, images, or audio output. It does this by finding and following patterns in the data on which it has been trained.
In contrast, agentic AI performs and executes tasks for users in response to real-time information and can make decisions on users’ behalf.
Example:
You might use GenAI to help brainstorm birthday present ideas for your 12-year-old cousin. If you ask it for a list of ideas, it would provide you with a customized and formatted list.
On the other hand, you could ask agentic AI to choose a present, find the best prices online, order it on your behalf (using credit card information stored in your browser), and have it delivered before the birthday party—all without your involvement.
Agentic AI browsers are based on large language models (LLMs), just like generative AI. Like generative AI chatbots, agentic AI browsers rely on training data to understand and respond to natural language commands from users. (See our online workshop about how generative AI and prompting works.)
So, in our example of buying a birthday present for your cousin, an agentic AI browser could use its training data to predict what might be a good birthday present for a 12-year-old. The more specific information you provide it, the better its purchase choice would be.
To actually pay for the present, the browser could use your stored credit card number or payment account—if you gave it permission to store and access that information. Currently, many agentic AI browsers will check with the user before completing purchases.
Agentic AI browsers can also incorporate other features like APIs, which allow them to request and receive information from websites to do tasks more efficiently.
Below are considerations to think about when using agentic AI for research or general use.
Privacy: Depending on how you want to use an agentic AI tool, it will require access to different types of data. This could include your email, work documents (like code, Word documents, or spreadsheets), your online calendar, your browser history, your stored payment information, or other private data that should be protected. Always read the data and privacy terms for any agentic AI tool you use, and make sure you understand how it will use and store your private information.
Examples:Lack of human control: Agentic AI tools can take initiative, make decisions, and act on your behalf. You need to maintain responsibility for monitoring an agentic AI tool’s actions and potential actions. Before you use an agentic AI tool, consider what the potential consequences would be if it took an action you didn’t intend or consent to.
Security concerns: Agentic AI is still a developing technology, and therefore its security is not guaranteed. If you use an agentic AI tool, be aware of the possibility of unauthorized access (e.g. hackers) or malicious data access, theft, or leaks.
For more digital security tips and considerations, check out the link below:
Ask us