In recent years, generative AI has dominated the news.
From Q&A and image generation, to coding whole applications, generative AI revolutionized the future of work and gave rise to an entire industry.
But it’s just beginning.
The next wave of the AI revolution is just around the corner with agentic AI.
Agentic AI will take AI capabilities to another level, further disrupting all areas of our lives.
This article takes a deeper look into what agentic AI is and what you should expect in the near future.
The foundation of agentic AI is built on top of generative AI models, so it’s important to understand the generative model before digging in further.
In simple words, generative AI is a large language model that excels at creating new data by learning from existing data.
For example, when an AI tool creates a picture for you, it doesn’t build it from scratch but based on millions of other pictures it was trained on.
This recreation process can range from:
Agentic AI goes beyond just responding to prompts and is all about action and autonomy.
It’s a system capable of observing its environment, creating a plan, and executing tasks independently to achieve predefined goals.
The most critical aspects of agentic AI include:
By collecting data from multiple sources, agentic AIs are able to understand the current “state of the environment.” It has real-time awareness of what’s going on around it and how the context of its work changes.
Based on predefined intent, agentic AI reacts to changes in context and sets relevant subgoals to achieve. It doesn’t need human input to direct it in the right direction or supervise its work.
Thanks to reinforced learning and other machine learning methods, agentic AIs improve their processes and way of working
Instead of merely generating content, agentic AI owns its workflow. It can search for information, schedule tasks, make decisions, adapt to new data, and handle unforeseen problems. Just like humans do.
Agentic AI works from five key parts:
Agentic AI isn’t a new type of LLM itself. Most agentic AIs are built on top of generative AI models.
For example, many existing agentic AIs are built on top of ChatGPT (generative AI).
The core model is enhanced by the policy layer — a set of instructions on how agentic AI should behave, including:
Most agentic AIs must remember past events to work effectively. This often requires storing hundreds of thousands of interactions, which requires appropriate storage capacity.
Agentic AIs go beyond displaying output in a chat form. They often require access to external APIs to get data from and push results to.
For example, an agentic AI that solves bugs might need access to Jira’s API to obtain ticket information, its own coding environment, and your CD pipeline to push bug fixes.
The agentic AI must be aware of whether the outcome it produces is positive or negative. It can be a machine learning algorithm, a human feedback system, or a combination of both.
The most fascinating part about agentic AIs is that they can do almost anything. Below are some of the best use cases:
As discussed earlier, an agentic AI could pull information from Jira tickets and based on the context of the existing code-base, fix bugs or even develop smaller features and push them to production with minimal human insight.
With access to real-time information about the state of inventory and needs for specific supplies, an agentic AI could order and even pay for new orders of goods to avoid interruption in production or order fulfillment.
With access to your emails, calendars, and to-do lists, agentic AI could serve you as a personal assistant, planning out your day and even booking airplane tickets.
While generative AI responses already greatly benefit customer support activities, AI agents can take them even further. Instead of just chatting with the customer and answering questions, agentic customer support agents could initiate refunds, change delivery addresses, or update account information on their own.
Yes and no.
Agentic AIs already outperform humans with tasks that are:
However, they lack human traits such as:
While an agentic AI could be an effective personal assistant, it will not chit-chat with your guests or help you unpack moral dilemmas like a human. The key is understanding its limitations.
So, yes, if your work is predictable, easily automatable, and requires little human interaction, agentic AI will most likely take over in the next couple of years. However, there will still be many places where human touch is irreplaceable.
There are two main ways you can leverage agentic AI as a product manager.
The first one is to automate your own work.
Are you summarizing meeting notes and sending them out to relevant stakeholders? Have AI do it. Or get AI to monitor your PRD files so that it can update Jira tickets for you.
Anything that doesn’t require skills or creativity can be automated.
The second one is to implement agentic AI in your product.
Look at the most painful customer problems and evaluate if they’re:
If yes, then they’re a good candidate for agentic AI.
Featured image source: IconScout
LogRocket identifies friction points in the user experience so you can make informed decisions about product and design changes that must happen to hit your goals.
With LogRocket, you can understand the scope of the issues affecting your product and prioritize the changes that need to be made. LogRocket simplifies workflows by allowing Engineering, Product, UX, and Design teams to work from the same data as you, eliminating any confusion about what needs to be done.
Get your teams on the same page — try LogRocket today.
Brian Fugere talks about the importance of having marketing, product, and UX constantly aligned early for a successful launch of any product.
Instead of needing to have a different app for each task, a super app lets you accomplish everything in a single place.
Let’s discuss why Skype is being discontinued, what could have been done differently, and a few important lessons for product managers.
Eugene Mandel talks about the importance of “show, don’t tell” with AI products, i.e., why users need visibility into AI applications.
Learn more →
