If you work in business, you’ve probably heard that artificial intelligence is going to revolutionize productivity in business operations. However, the limitations of commercially available artificial intelligence, namely large language models (LLMs), prevent businesses from realizing the full potential of this powerful technology. Instead, the next step in the revolution of AI in business is agentic AI, which promises to bring us one step closer to full automation with custom solutions that actually help companies to achieve their operational goals.
Agentic AI is an evolution of generative AI technology that is available on a broad scale, but is notable for its ability to exhibit autonomy, goal-driven behavior, and adaptability. The primary goal of agentic AI models is to make intelligent decisions based on the information available to them, allowing them to solve complex problems in real time and act independently with minimal human supervision.
Why agentic AI is the future of AI for business
Indeed, agentic AI is different from consumer-facing LLMs like ChatGPT or Google Gemini, which most users are likely familiar with. LLMs are just transformers that predict the next set of tokens broadly based on a prompt using information from the data the model is trained on and context from the prompt. On the other hand, agentic AI is designed to accomplish specific tasks and workflows in a fully automated fashion (or as close to fully automated as possible).
“Agentic AI systems are intelligent goal-driven systems,” explains Mridul Nagpal, founder and CTO of AI solutions provider Krazimo. “They typically operate by creating a plan to accomplish the task given to it which can include calling on tools that can 1) Get access to additional information that’s not available to the LLM because it’s proprietary or because it’s from live data, more popularly known as RAG or Retrieval-Augmented Generation; or 2) perform actions like sending emails, uploading files, executing transactions, etc. They typically iterate autonomously until a task is done — escalating to humans only when necessary.”
Nagpal’s company, Krazimo, specializes in creating curated, enterprise-grade AI solutions for businesses that help boost employee productivity and supercharge organizational growth. As part of the company’s mission to help businesses remain competitive in an evolving landscape, Krazimo is dedicated to helping its clients explore the latest in artificial intelligence and machine learning technologies, which currently comprises agentic AI.
Businesses may wonder how they can leverage the power of agentic AI to improve productivity and efficiency in their operations. Nagpal points to an example of responding to email inquiries to illustrate the potential of agentic AI solutions:
“In an LLM workflow, a human employee would have to open the email, find the service description in their documentation, put the service description and the pricing structure in an LLM along with the original email, ask it to generate a response, fine-tune the response, and copy the response back into the email,” he explains. “An agentic AI does all the steps above as soon as the email comes in, including fetching the required internal documentation about the service, and asks the human to validate the email before it sends it (or if it’s trusted and the stakes are low, just sends it).”
How companies can use AI solutions to support their business needs
While that “just send it” level of full automation is clearly the ideal for businesses that hope to optimize their efficiency, the truth is that it takes time and an effective strategy for an agentic AI product to boast this trustworthiness. Thus, the question becomes how to integrate agentic AI effectively into workflows.
Nagpal suggests that inserting determinism at every possible step is the best approach. Additionally, a strategy known as “human in the loop,” in which a human employee validates an AI agent’s outputs until it reaches a certain performance threshold at which it can become fully autonomous, is invaluable in launching reliable AI agents.
“Split the actions taken by LLMs into modular tasks that can be evaluated and tested separately,” Nagpal suggests. “Add individual agents for each functionality and build adequate test suites around their function calls, giving them restricted access to resources, just like people working in an office with different roles.”
Those entrepreneurs hoping to harness the full potential of AI for business should look beyond the broadly available LLMs and toward the power of custom AI solutions that include agentic AI tools. By embracing the full automation of agentic AI, businesses can be part of the AI revolution and reap the benefits of greater productivity and efficiency.
“I believe that in two years, agentic AI will be capable of replacing anywhere from 40-70% of the tasks that knowledge workers do today,” concludes Nagpal. “The low-hanging fruit is everything that people are currently copy-pasting into ChatGPT, but many more complex multistep processes will also be automatable. There are obvious applications, such as customer service, outbound sales, company training, and in-house data scientists/report generators, but I think it could go well beyond that. For example, Cursor is already acting as a blazingly fast junior developer. I believe it’s possible that the first draft of many of the most complex problems we tackle at work today will be done with AI.”
link