How Big, Old Companies Could Win the Agentic AI Race
Large enterprises with messy tech stacks can win if they are willing to fund the infrastructure that makes AI agents safe, scalable, and worth trusting
📌 THE POINT IS: Big, old companies are not too late for agentic AI. They may actually have the most to gain, if technology leaders are willing to invest right. The agentic race will be won by technology leaders in large enterprises who fund the infrastructure that lets AI scale without breaking trust. Do that well, and the payoff is bigger than efficiency alone: it’s the chance to redeploy talent, improve service, and build the next version of the company.
Why big companies feel behind
You’re the CTO of a large company. Every day you wake up to news about AI agents taking over workflows, working all night on projects, and shipping code at never-heard-of speeds. Then you bury your head in your hands wondering, “Why can’t I have this life!?” Most of the incredible news that you hear about these days that tout large, agentic wins are companies that were built on digital rails (e.g. newer, usually smaller ones).
No wonder you’re feeling frustrated and behind…their realities are very different than yours.
“More than half of companies (51%) have already deployed AI agents at their organization and another 35% plan to deploy within the next two years.” — PagerDuty, 2025 survey of 1,000 IT and business executives
I’ve lived the mess
For most of my career, I’ve worked at large, old enterprises. GE and Bank of America were > 100 years old when I worked at each and now Nationwide is celebrating its 100th year of protecting people, businesses and futures with extraordinary care. As large-sized corporations, they all have diverse technologies that have either been bought, built, or acquired through purchases of other companies. The technology environments are old, vast, complicated, and most of the time held together with digital bubblegum and duct tape.
This is not the environment in which you let AI agents (let alone humans!) run free. There are decades of crazy code written well before CI/CD automation and testing were ever a thing embedded in some of the most critical systems that keep the place going! In a large enterprise, bad automation does not just break a toy process. It can break customer trust, compliance posture, or core operations.
“Nearly two-thirds of enterprises worldwide have experimented with agents, but fewer than 10 percent have scaled them to deliver tangible value.” — McKinsey, 2026
Don’t freeze. Build the stack.
I’m here to tell you that while all of this is true, there is no reason to fear or freeze in your tracks when considering starting with AI agents. I’ve talked a bit about Knowledge Graphs and Context Graphs as early steps to getting your data environment ready for AI. There are a number of other “infrastructure” components such as AI gateways, orchestration tools, and model selection and evaluation wizards that you’re just going to have to buckle down and implement. Deloitte found that while 42% of companies felt strategically ready for AI, they felt less prepared on infrastructure, data, risk, and talent.
AI isn’t going to be an overnight project for you like it is for the Ma-and-Pa shops out there that aren’t carrying the burden of providing electricity to homes, facilitating everyday people’s wallets, or protecting your family’s home. I hate to say it, but the large, old companies out there in many cases are providing the most critical elements that we all need to live safely. There’s just more responsibility on your hands and that means you need to spend more money to get there.
The ROI story that most leaders get wrong
But it’s the ROI upside that you have to keep your eyes on. The largest failing that technology and business leaders have today is that they think that ROI should come in the form of decreased costs and a wider margin on the bottom line. That’s a good secondary goal to have, but the real goal should be growing your business, increasing your top line through offering new products and services, and finding ways to reimagine your company for its next 100 years. Having AI process call center transactions will improve your margin, sure, but if you deploy the humans who were spending days answering phones and train them to talk to those same clients to uncover potential new business offerings or better ways to offer value to your clients, you’ll see an even larger overall impact.
Let’s take the very recent example of American Airlines. The airline is launching electronic boarding gates at DFW to streamline boarding and free up gate agents to help customers with more complex issues instead of focusing on the repetitive task of monitoring boarding-pass scans. That is the kind of “boring” use case that scales in a large enterprise: automate the simple, repeatable work so your people can spend more time on higher-value service and operational needs. Over time, that kind of change can give operations leaders more flexibility in how they deploy staff during disruptions, delays, and high-touch customer moments.
“Streamlining manual tasks for American’s team members during boarding, allowing them more time to provide exceptional customer service and operationally critical tasks.” — American Airlines press release on its electronic boarding gates at DFW
5 things you can start with today
So how do you get started if you’re staring at a valley of mainframes, COBOL code, and tears? Here are your first five steps:
Pick one ugly, high-value workflow and map it end to end. Don’t start with a moonshot. Start with a process that is painful, repetitive, and economically meaningful...like American Airlines!
Identify the systems of record, data sources, and decision points that workflow touches. If you do not know where the context lives, your agent will not know what “right” looks like either.
Put guardrails in before autonomy. Define who approves what, where humans stay in the loop, what actions require escalation, and how logging and audit trails will work.
Build the context layer on purpose. That may mean a Knowledge Graph, Context Graph, memory layer, or all three. Whatever you call it, your agents need business meaning, not just tokens.
Track value beyond labor savings. Measure cycle time, quality, customer impact, conversion, risk reduction, and the new revenue opportunities your people can pursue once AI takes the grunt work off their plates.
Act your size, but not your age!
If you’re a large, old company, this is the part where you stop envying the startups and start acting like the institution you are. Your edge is not speed alone. Your edge is trust, scale, customer reach, and the ability to reinvent critical work in ways that matter. The winners in this next era will not be the companies that rushed agents into production with the fewest guardrails. They will be the ones that invested early in the infrastructure, context, and governance required to scale AI without blowing up the business. That’s how old companies will continue to get older…and better!