Jay Bhatty is the CEO and founder of NatGasHub.com.gettyJust a few years ago, it would have seemed like sci-fi to imagine people using AI regularly at work. But here we are. ​LLM and GenAI are now commonly understood terms. A February 2026 Gallup survey found that "half of employed American adults say they use AI in their role at least a few times a year," and 28% say "they use it a few times a week or more," a jump from 46% and 26%, respectively, in Q4 2025. ​After its Next 2026 conference, Google Cloud published a piece saying, “the most significant trend is the transition from AI as a passive assistant to AI as an active part of the team, where specialized agents can orchestrate entire workflows.” ​This is the trend I’m seeing unfold in the energy sector. It’s clear to me that agentic AI is the future of work. My company started developing agentic AI software for clients in the oil and natural gas industry less than two years ago. Our goal is to build up to 10,000 AI agents for clients working across different natural gas pipelines in the next 12 months. From LLMs To AI Agents To Agentic AI FrameworksLLMs are quickly evolving into commodities, like any other type of technology. You may have loyalty to one brand over another, but each one deals with the same limitation: Its training data has a cutoff date.​If you ask an LLM, "What is tomorrow's weather looking like?" or "What are the average airline ticket prices to Boston next week?" it doesn't have that information. An LLM comes out of the box like a laptop, calibrated with an operating system trained on data with a certain release date.​You can take an LLM to the next level by combining it with a real-time source of data. This is where AI agents come in. Picture an agent as a human employee with a body (software instructions) and a brain (an LLM). ​Ideally, the agent will follow your instructions with no issues: “Go to Website A and compile all the airline ticket prices for these dates. Then go to Website B and look at the weather for those dates. Then book the tickets for dates when the weather is predicted to be good and the tickets are below a certain price.”​If it does get stuck, it can ask the LLM for guidance: “The weather on these two days is the same; which would you prefer?” The LLM is continuously learning. You can train it to react in prescribed ways to different situations. If this happens, then do this. ​Once you expand from one AI agent to multiple agents, you’ve now built an agentic AI framework. You can link these agents to one or more LLMs, weighing the pros and cons of comparing answers from different brains and spending tokens for each tool. Lessons In Expanding Agentic AI 1. Treat AI Agents Like New EmployeesYou could hire a brilliant Harvard or MIT graduate, but on their first day on the job, they would still need to learn about your company, your culture and your products. It’s the same concept with an AI agent. Give them clear instructions so they can successfully sell your products. ​A new employee might ask you, “How do I deal with this situation with this client?” and spark a teaching moment. While an employee may make the same mistake twice, an AI agent won’t. If you code instructions for each scenario you encounter, the agent will respond by following those precise steps. Keep updating the instructions, and your agents will keep learning. ​You also need to train your LLM to be a reliable knowledge base for the agent to query. The LLM doesn’t come out of the package knowing anything about your company, clients, revenue or sales. Train it to understand the nuances of your business. When I’m building AI software for clients in the energy industry, I need to train LLMs on industry-specific compliance guidelines, SKU numbers and pipeline receipt and delivery locations. 2. Act Like A ManagerThe way I look at my business model, I’m a human supervisor managing an assembly line of thousands of AI agents. My job is to build technology that automatically monitors all of these agents to perform their tasks to completion and deliver the results to the client's website. ​Accept that an agentic AI framework will change your workforce. There will be jobs lost, but there will also be jobs created. Get comfortable managing both human and digital workers, in whatever ratio works for your business. 3. Focus On Repeatable Workflows What tasks do your employees dread that they have to repeat every day? This is where you can benefit most from agentic AI. Immediate benefits include reduced errors, costs, headcount and pressure around deadlines. The biggest secondary benefit relates to opportunity cost. How many hours a week are people spending on these repetitive workflows? How could they use that time for revenue-generating or innovative ideas? ​When I started building agentic AI software for clients, I focused on three workflows that are common to shippers across the entire energy industry: pipeline nominations, tariffs and invoices. Since every company that moves gas from point A to point B on the North American network has to repeat the same workflows, they were good candidates for automation. 4. Look For Industry Expertise Deciding whether to build or buy agentic AI software can be a stumbling block. There’s a misconception that it’s so easy to code now that any company can build its own AI agents and save money. But you need to factor in CapEx as well as OpEx—your initial investment in agentic AI, plus the expenses associated with maintenance year after year. Even tech companies often use the tools of other tech companies—email or cloud storage applications, for example—so they can focus on their core business. ​A software partner that has expertise in your specific industry can distribute costs among their clients, including your competitors. In the energy industry, for example, software providers need to build AI agents to meet certain rules and regulations. My company manages compliance, security and privacy laws for all our clients, so none of them has to do this internally. ​At the end of the day, agentic AI software is just another software solution. Today, you may need an AI agent to automate your front office. Tomorrow, you may need one to automate the back office. When you’re hiring human employees, you probably want to work with one recruiter for all open positions. The same idea applies to an AI software partner. Look for one vendor that understands your industry and can handle all the different workflows within your company.​Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?