Agentic AI vs. generative AI
Both agentic AI and generative AI (gen AI) offer productivity benefits by assisting, augmenting, and optimizing tasks and processes. Both are forms of artificial intelligence that use large language models (LLMs).
When comparing the 2, think of agentic AI as proactive and gen AI as reactive.
- Agentic AI is a system that can proactively set and complete goals with minimal human oversight. If part of accomplishing that goal involves creating content, gen AI tools handle that task. Agentic AI becomes an agent of the user and/or system.
- Gen AI is a tool that creates new content in reaction to a prompt. It can be a component of an agentic system, but it doesn’t act on its own to complete a task. It has no agency.
Agentic AI and gen AI do work collaboratively. Agentic AI systems may use gen AI to converse with a user, independently create content as part of a greater goal, or communicate with external tools. In other words, gen AI is a critical part of agentic AI’s “cognitive” process.
What are the differences between agentic AI and generative AI?
The line between agentic AI and gen AI can feel blurry because they both begin with a prompt from a user and typically exist in a chatbot-like format. Plus, many applications that were once purely gen AI now include agentic elements—a trend that’ll likely continue.
For example, many of the popular chatbot platforms (ChatGPT, Gemini, Claude, etc.) automatically initiate a web search, parse the data, and return it as part of the conversation. This is a primitive form of agentic AI.
Agentic AI and gen AI differ in their ability to act independently and collaborate with external tools
To better explain the difference between the 2 technologies, let’s explore a hypothetical use case:
A sales representative wants to use AI to write a follow-up email to a sales lead.
With generative AI, the sales representative would open a gen AI interface and type a prompt like, “Write a polite and professional follow-up email to Maria Wang about our proposal.” The gen AI instantly produces a draft of the email and has fulfilled its purpose. It’s now up to the sales representative to copy that text, paste it into an email, enter the recipient’s email address, and hit send.
Now let’s explore how agentic AI would handle a similar task.
Within an agentic system, the sales representative would set a rule or command in their customer relationship management (CRM) system. It might say something like, “For any sales lead I mark as ‘Follow-up required,’ wait 2 business days, then send a follow-up email.”
Once the sales representative marks Maria Wang as “Follow-up required,” the agentic workflow is triggered. The system has its instructions (the initial prompt) and independently lays out a plan to enact with the help of external tools. The plan may look something like this:
1) After 2 business days, the system sends a request to the agentic workflow.
2) The system retrieves Maria’s details from the CRM.
3) Another tool fetches additional information about Maria (customer history, personalization details, company information, etc.) that provides context for the prompt for the follow-up email.
4) The system creates a prompt for the follow-up email and provides it to an integrated gen AI model, which writes the email text.
5) The system provides a draft of the follow-up email to the sales representative, who approves the email or sends it back to redraft.
6) If the sales representative approves the email, the system makes an application programming interface (API) call to Maria’s email service.
7) The system sends the email to Maria.
8) The system updates the CRM to show the email has been sent.
Agentic AI and gen AI differ in their ability to adapt
Both types of AI are adaptable in their own way. Generative AI expresses its adaptability by producing content in many styles and for different contexts. Agentic AI showcases its adaptability by adjusting its plan and strategy in response to changing environmental conditions or new information.
Agentic AI provides a framework for workloads that used to be classified as robotic process automation (RPA). The injection of AI makes agents far more adaptable to changes in their runtime environment. For example, screen-scraping bots would struggle with even minor changes on the target site, whereas agentic AI can adapt to changes and adjust its approach to harvesting data. Therefore, AI-powered agents can operate on a level that previously would've required human input.
Context understanding
Agentic AI uses context for action. Agentic systems have access to:
- The initial prompt.
- The state or conditions of the digital or physical environment.
- Available tools (API access, a gen AI application, etc.).
- Memory and past actions.
With all this information, plus mathematical formulas that help it navigate the data, an agentic system gains an understanding of context. This is what allows it to “reason” and take action.
Meanwhile, generative AI uses context for creation. Because the goal of gen AI is to produce new content, generative applications have access to:
- The prompt.
- Conversational history.
- Data they were trained on.
With this information, plus machine learning techniques and deep learning algorithms, gen AI can draw information, create connections, and produce an output.
4 key considerations for implementing AI technology
What is agentic AI?
Agentic AI is a software system designed to interact with data and tools in a way that requires minimal human intervention. With an emphasis on goal-oriented behavior, agentic AI can accomplish tasks by creating steps and performing them autonomously. Agentic AI can set its own goals, delegate tasks to other AI agents or external tools, and adapt to new or unstructured conditions that it wasn’t trained on.
AI agents are components within an agentic system. Think of an AI agent as an entity that sits on top of other software tools and operates them. Agentic AI can be a physical structure, a software program, or a combination of the 2.
An AI agent in a robotic system might use cameras, sensors, and monitors to collect data about its environment, then run that information alongside software to determine its next step. A good example of this would be an autonomous vehicle encountering debris on the road and deciding whether to press the brakes or keep driving forward.
Meanwhile, agentic AI in a software setting would collect data from other sources—such as APIs, online searches, text prompts, and databases—that help the agents create a sense of perception and context. Consider our previous example of the employee who wants to automate the multistep task of sending out a personalized follow-up email after meeting with a potential client.
How does agentic AI work?
Agentic AI can solve problems by following 3 steps: perceiving, planning, and acting. It has a “chaining” ability, which means it can perform a sequence of actions in response to a single request or prompt.
For example, if you ask an AI agent to create a website, it can perform all the necessary steps. This means that from 1 prompt, the AI agent can write the code for the structure, populate the pages with content, design the visuals, and test for responsiveness.
In this way, agentic AI can be thought of as a “doer” and a “project manager.” It can navigate any hurdles it runs into and initiate action, such as creating its own prompts to help answer questions that arise.
What is an agentic workflow?
Agentic AI works because of a process known as an agentic workflow. An agentic workflow can be made up of an orchestration of agents, robots, and people. It’s an end-to-end process designed to achieve a specific goal. It bridges the gap between digital and physical worlds while integrating human oversight.
An agentic workflow is a structured series of actions AI agents manage and complete, sometimes with a human in the loop. When an AI agent is given a goal to complete, it begins the workflow by breaking down a task into smaller steps and then performs those steps.
To carry out these steps, an AI agent spins up more versions of itself, creating a multi-agent system (MAS). In this workflow, the main agent (also known as a meta agent, orchestrator, or supervisor) may create new agents and delegate tasks to them, assigning values and interacting with memory in a feedback loop. The agents work in parallel until they reach the overall goal.
Within this MAS, each agent is made up of an internal structure that allows it to function both independently and collaboratively within its system. This collaboration is dependent on shared memory stores, which provide context regarding individual knowledge, past experiences, and belief states.
Agentic AI use cases
Agentic AI excels in dynamic problem-solving and decision-making. Some industry-specific use cases for agentic AI include:
Manufacturing: Agentic workflows can help manage supply chains, optimize inventory levels, forecast demands, and plan logistics.
Healthcare: Agentic AI can engage with clients by monitoring needs, carrying out treatment plans, and providing personalized support.
Software development: Agentic AI can automatically generate debugging code, manage development lifecycles, and design system architecture.
Personalized employee support: Agentic AI can adapt its approach as situations change and offer tailored and proactive support. This means it can help complete tasks like scheduling, answering questions, and onboarding.
Financial risk management: Agentic AI can assist in finance and trading through its ability to analyze market trends, make trading decisions, and adjust strategy based on streams of real-time data.
What is generative AI?
Generative AI is a type of artificial intelligence that can produce new content, such as text, video, audio, and software code. Gen AI uses deep learning to calculate statistical relationships between words and creates output based on its training data, pattern recognition, and probability.
Generative AI is reactive, meaning it must be prompted with a specific query before generating a response. It can’t set its own goals, delegate tasks, or adapt to new or unstructured conditions.
The content a gen AI application can create is limited to the data it was trained on. However, you can use techniques like retrieval-augmented generation (RAG), which incorporate external data sources, to make the output of gen AI models more accurate.
Generative AI use cases
Generative AI primarily supports human decision-making by providing information and options a user can take or leave. Some use cases for gen AI include:
Writing: Gen AI tools can respond to prompts for written content creation on practically any topic. These tools can also adapt their writing to different lengths and writing styles.
Image and video generation: Gen AI image tools can create high-quality pictures and add new elements to existing works. Many gen AI applications also offer tools that can whip up a short video in response to a prompt.
Speech and music generation: Using written text and sample audio of a person’s voice, AI vocal tools can create narration or singing that mimic the sounds of real humans. Other tools can create artificial music from prompts or samples.
Code generation and completion: Some generative AI tools can take a written prompt and output computer code on request to assist software developers.
Data augmentation: Generative AI can create a large amount of synthetic data when using real data is impossible or not preferable. Synthetic data can be useful if you want to train a model to understand confidential data without including any personally identifiable information. You can also use it to stretch a small or incomplete data set into a larger set of synthetic data for training or testing purposes.
Considerations for governance, safety, and trust
Perhaps the most common concern when considering agentic AI is: Who is responsible when an autonomous system makes a mistake? In other words, how do you balance autonomy and oversight?
Human-AI collaboration
Before implementing an agentic system, it’s important to create a framework for accountability, transparency, and control.
Consider our previous example of the sales representative using AI to send an email to a potential client: The employee would likely want to see the email prior to sending it.
Agentic AI can make decisions independently and with minimal human input. But that means you’re trading efficiency for complete oversight. One solution is to focus resources on testing and validation. This means keeping a human-in-the-loop mindset to monitor actions and prevent poor judgement calls.
Similar considerations for trust and safety apply when implementing gen AI systems. Perhaps the most obvious risk gen AI poses is the ability to create misinformation or disinformation. This includes instances of perpetuating harmful biases and stereotypes as well as the creation of deepfake images for malicious intent. It’s important to be cautious of “hallucinations” (inaccurate outputs that present as factual) and fact-check rather than taking answers at face value.
Privacy and security
Agentic AI’s ability to reach into external databases creates more opportunities for security and privacy risks. This means you need security frameworks that safeguard the data coming in and out of your workflow.
Generative AI can also pose security risks. Users may enter sensitive information into applications that lack security. Gen AI can also introduce legal risks by reproducing copyrighted material or appropriating a person’s voice or identity without their consent.
How Red Hat can help
Red Hat® AI is our portfolio of AI products built on solutions our customers already trust. This foundation helps our products remain reliable, flexible, and scalable.
Red Hat AI can help organizations:
- Adopt and innovate with AI.
- Break down the complexities of delivering AI solutions.
- Deploy anywhere.
A foundation to keep your options open
Red Hat AI solutions can support both generative and predictive AI capabilities. With the flexibility to bring your own model, Red Hat offers support for training and fine-tuning foundation models specific to your business use case.
A good place to start is Red Hat Enterprise Linux® AI, a foundation model platform that helps you develop, test, and run LLMs and SLMs for enterprise applications. The AI platform gives developers quick access to a single-server environment, complete with LLMs and AI tooling. It provides everything you need to tune models and build gen AI applications.
From there, explore Red Hat OpenShift® AI, which provides a unified platform to create multiagent systems. Plus, you can control the adaptive learning and reasoning AI agents use via Red Hat OpenShift’s MLOps capabilities.
Additionally, our AI partner ecosystem is growing. A variety of technology partners are working with Red Hat to certify operability with Red Hat AI. This way, you can keep your options open.
Artificial intelligence (AI)
See how our platforms free customers to run AI workloads and models anywhere.
