The Agentic Edge: Who Has It and How to Get It

For anyone under 30, here’s how it worked. You walked into a little storefront with a poster of a Caribbean island in the window and sat across the desk from someone with a phone wedged on one shoulder, typing into a green screen. It took an hour, maybe more, to square away all of the details, calling all of the airlines and hotels to get availability and secure the reservations. A week or so later, your paper airline tickets showed up in the mail in a little envelope, and you guarded them with your life, because if you lost them, that was the end of that trip. Here’s a little clip from My Big Fat Greek Wedding of what travel agents looked like “back in the day.”

Then the internet came along, and we all became our own travel agents. Storefronts with posters became smoothie shops. We spent nights with six browser tabs open, comparing fares and clearing our cookies because somebody swore the price went up if you looked twice.

Over time, the travel sites got smart, quite literally. They learned what we liked, guessed when fares would drop and nudged us toward the hotel we’d probably book anyway. That’s AI, and it’s been helping us plan trips for years.

But for all of its smarts, it never booked the trip. It made suggestions. You still did the work. You picked the flight, typed in your passport number for the millionth time and spent an hour on hold when the flight got canceled.

Now that’s starting to change. AI agents are learning to take a request like “Get me to Chicago Tuesday morning, aisle seat, home by Thursday night,” find the options, book them and rebook you when a storm rolls in.

Read More: Agentic Commerce Needs Its Febreze Moment

That isn’t a smarter search engine. It’s a different job. The question changes from “What does AI suggest?” to “Who’s doing the messy work of the booking?”

The agent is back. It just isn’t sitting behind a desk anymore with a headset.

Big companies are at that exact moment with AI. The ones putting AI and agents deepest into the work are pulling ahead. We call that the agentic edge. Who has it? You might be surprised.

AI isn’t new. Agents are.

For all the noise about AI, it isn’t new to big companies. Banks and card issuers have used machine learning to spot fraud and stop cyberattacks for decades. Retailers use it to forecast demand. Airlines use it to price seats. Insurers and lenders use it to size up risk. That AI is already inside the business, and it’s very good at its job.

But that AI makes suggestions. It tells people where to look, what to charge and which transaction seems off. People still make the call and do the work. And still hopscotch across the organization to pull together all of the relevant information to make the report that drives the decision. It takes hours, days, sometimes even weeks.

GenAI added a helpful assistant. Chatbots draft the email, summarize the meeting and clean up the slide. Useful? Absolutely. But a person still does the job.

AI agents are different. An agent doesn’t just suggest what to do next. It does it. It reads the invoice, matches it to the purchase order, pays the ones that check out and sends the rest to a person. It looks at the email and writes a response and then organizes the to-do list and sends relevant emails to those who need to do stuff. It works the case from start to finish while a person checks its work.

But here’s the thing. If every AI agent disappeared tomorrow morning, most big companies would be fine. Sort of. People would go back to doing the work themselves, and the business would run. People would be more grumpy, there’d be more friction, but it’s not as if businesses and the economy would come to a grinding halt.

That’s how new agents are. It’s also why this is the moment that matters. Agents are where the next big gains in speed, cost and growth are likely to come from, and the companies running AI and agents deepest are already pulling away.

Read More: Karen Webster Says the Agentic Enterprise Is Already Here

The conversation has shifted from AI and the enterprise to AI inside the enterprise. The Agentic Enterprise, to be exact. But “Agentic Enterprise” is a talking point without a clear definition.

So, the team at PYMNTS Intelligence has built a way to measure it, and a framework businesses can use to see how and where AI and agents are driving measurable business outcomes. In other words, who has the agentic edge. And before you read too much further, know that no company has cornered the agentic edge. In fact, enterprise companies didn’t fare well in our initial framework.

Asking the Wrong Questions About AI

Since March 2024, PYMNTS Intelligence has surveyed executives at U.S. companies with $1 billion or more in revenue nearly every month about the use, adoption and returns of GenAI and, more recently, AI agents. That’s 27 surveys and 1,627 executives, from CFOs and COOs to the people running AI strategy at some of the biggest companies in the country.

For most of that time, the market has been asking one question of them. Are you using AI?

That question is now mostly a non-question. The answer is yes, and for a lot of companies it’s been yes for years. In our most recent data, AI shows up in 98% of the business areas we track. And 97% of executives say it’s paying off, including 94% of the companies that are furthest behind.

When everyone gets the same grade, the test isn’t telling you much.

Read More: Gen AI: The Technology That Broke the Adoption Curve

The question that matters the most now is how much of the work AI and agents actually do. Is it still making suggestions while people do the job? Or have agents taken over real work inside the processes that make the company money and defend its core assets?

We’ve found that you can’t answer those questions with a count of use cases. A company with 200 chatbots and a company with agents working its fraud cases don’t look the same, and they shouldn’t score the same.

Five Rungs to Measure Agentic Enterprise, and the Line That Matters

We’ve created the Agentic Edge Index to score the degree to which enterprises are really agentic. It measures how far AI and agents have moved from helping people to doing the work in 75 activities across eight parts of the business, from fraud detection and treasury to pricing, supply chain, HR and customer service.

For each activity, we look at where AI sits on a five-rung ladder:

1. Chatbots and GenAI tools are being tried. A pilot here, a one-off there.
2. Chatbots and GenAI copilots regularly help people do the work.
3. Chatbots and GenAI move from sidecars to being built into how the work gets done every day, but people still do it.
4. AI agents carry out the work, with humans in the loop to sign off.
5. AI agents run the routine work on their own, and people handle the exceptions.

The first three rungs are chatbots and GenAI tools that help people. The top two rungs are agents doing the work. That’s the agent booking the trip.

The jump from the third rung to the fourth is what we call the agent line. Below it, people do the work. Above it, AI does.

Then we do something many AI scorecards don’t. We weight the activities by how much they matter.

Read More: Why Measuring the ROI of Transformative Technology Like GenAI Is So Hard

Not all AI is created equal. AI that helps analyze customer survey comments is nice. AI that decides which transactions look like fraud, or which security alerts need a human right now, is a different animal. If it gets it right, it protects revenue and reputation. If it gets it wrong, it costs real money and sometimes makes headlines.

How We Weight What Matters

Every one of the 75 activities we assembled and track is rated from 1 to 5 for its business impact and for the cost of getting it wrong. The work that matters most to the business counts most in the score. A company that has agents working inside its high-stakes work will score higher than one that has sprinkled chatbots across a hundred low-stakes tasks, and it should.

The result is one number from 0 to 100. The score reflects how much of the work AI supports and how far up the ladder that AI has climbed. A 100 would mean agents running every activity we track, across the whole business, on their own. A typical company, with AI in every part of the business, mostly in limited use, and agents in only a few areas, scores about 15. Getting to 50 takes agents at work in about half of the business, mostly built into the daily work and acting with a person’s approval.

Read More: The Battle for AI Isn’t About Models. It’s About Habits

It’s a simple way to answer the question at the top of this piece. Who’s doing the work, and is it the part of the business that counts?

In other words, we measure how far along any business is on the road to becoming an agentic enterprise, and how much of the agentic edge it has.

Agents Went From Locked Out to Let In

So where are America’s biggest companies on that ladder?

Mostly below the agent line. The typical enterprise scores in the teens out of 100. AI is everywhere, but mostly as a helper.

What’s changed is how fast companies have opened the door to agents. In late 2025, just 7% of the executives we studied were using or piloting AI agents, and 67% said they wouldn’t give agents any access to company systems at all.

By 2026, among the AI decision-makers we surveyed, 89% were using agents in at least one part of the business. Only 2% would refuse them any access. It’s a stunning shift in a very short time.

Here’s the fine print. Most companies have let agents in the door but haven’t let them do much yet. Seventy-eight percent would let agents read company systems but not act on them. Another 18% would let agents carry out tasks with a human’s approval. Just 2% would let agents run routine work on their own.

In other words, the agents can look up the flights, but they can’t book them. That’s the gap. It’s also both the opportunity and the roadmap.

Agents Start Where AI Has Been Present the Longest

Now take a look at where agents are being trusted to act, and a pattern jumps out.

Of the 75 activities we track, 17 are both high-risk and high-impact. These are the jobs where a mistake costs real money and has the potential to create a negative consequence to the business. Fraud and financial-crime detection; regulatory compliance; credit risk; security monitoring; treasury; payment routing; the monthly, quarterly and year-end close.

AI shows up most in exactly those jobs. On average, 44% of enterprises use AI for each of those high-stakes activities, compared with 39% for the lower-stakes work. Security monitoring leads all 75 activities at 55%. Regulatory compliance monitoring is at 51%. Fraud and financial-crime detection is at 50%.

Read More: The AI You’re Given vs. the AI You Chose

What’s fascinating, maybe even a little counterintuitive, is that those highest risk and impact activities are where the companies at the top of our index lean in hardest. They use AI in half of them, compared with 39% for companies in the bottom third. And when they use AI there, it’s 75% more likely to be in a part of the business that runs AI agents, 45% vs. 26%. Waiting is the bigger risk. Attackers work in real time, so the defense has to as well.

Agents themselves are still early. Companies would let agents act in about one in ten of the cases where they use AI, and that share is about the same for high-stakes and lower-stakes work. Among the 23 high-risk activities we track, 11 have AI agents taking action in at least 10% of the cases where AI is used. Nine of those 11 are security, fraud, risk or governance jobs. Security monitoring and data governance lead, with agents acting in about 16% of use cases.

It makes sense when you peel back the layers. Across the industry, these are likely the teams that have typically run machine learning the longest. They live comfortably and confidently in a world of models, rules, thresholds and audit trails. They know how to measure whether AI gets it right. So when agents came along, it’s no surprise they were among the first ready to hand over part of the job.

And in those jobs, speed is the whole game. Fraud doesn’t wait for a meeting. A cyberattack doesn’t wait for a human to read the alert. Agents that can act in seconds, not hours, are worth a lot there.

So the story isn’t that AI is new. It’s that the teams with the most AI experience are likely to be among the first to let it do the work, the farthest along on the road to being a truly Agentic Enterprise.

The rest of the business world is about to follow them.

What Goes With the Agentic Edge

Here’s where the rubber really meets the road: the balance sheet.

When we split companies into thirds by how deeply AI and agents run inside them, the top third doesn’t just feel better about AI and agents. It’s also growing faster.

Companies in the top third are 97% more likely to have grown revenue by more than 15% than companies in the bottom third: 24% vs. 12%. They’re also 97% more likely to have grown margins by more than 5%. And 82% of them grew both revenue and margins, compared with 66% of the bottom third.

Read More: How Leading Enterprises Really Measure Gen AI ROI

The relationship holds across the whole range, not just at the extremes. Every 10 points on the index goes with more than double the odds that a company grew revenue by more than 15%. Leaders are also 34% more likely to call their returns very positive, 64% vs. 48%. And while nearly every company reports a positive return on AI, it’s the leaders who are seeing it show up in faster revenue and margin growth.

The leaders are the ones putting agents to work. They use agentic AI in 41% more of their business areas than the bottom third, 42% vs. 30%, and they’re more than twice as likely to give agents real room to act, 28% vs. 12%.

One of the biggest gaps is in what AI actually does. In the top third, eight in ten enterprises say AI has replaced manual work, either in some areas or across most of them. In the bottom third, it’s 45%, little more than half that. A 10-point step up the index, from 15 to 25, goes with companies being 48% more likely to say AI and agents have replaced the time-consuming drudgery of manual work, 86% vs. 58%.

That’s the agentic edge, and here’s why it could matter. When AI helps people, it saves a few minutes here and there, and the savings disappear into the day. When agents do the work, they can take whole steps out of the process. Five steps become two on the way to one. That’s how the savings could show up in the margin, and the speed in the top line as growth, without needing headcount to support it.

Read More: The Big GenAI Time and Money Shift

Now, this is an association, not a controlled experiment. Deeper AI and faster growth go together, and the data can’t tell us which comes first. Deeper AI may help companies grow faster, faster-growing companies may be quicker to adopt agents, or both.

But size alone isn’t what sets the leaders apart. Profitability and conviction do. They’re more than twice as likely to earn net margins above 10%, 35% vs. 17%. Leaders are more likely to call AI a fundamental shift for their industry, and far more likely to say its benefits outweigh its ethical and other concerns. They’re willing to change how the work gets done to make it pay. And they’re living it, putting AI and agents to work instead of just talking about them.

Being big can help you pay for AI. It doesn’t always make you use it well.

How the Leaders Get the Agentic Edge

So what are the companies at the top doing differently? It isn’t one big bet. It’s a handful of choices that add up.

We’d love to be your preferred source for news.

Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!

Add as Preferred Source

They let agents do the work. This is the big one. Leaders are far more willing to let agents carry out the work, with a person approving it, instead of keeping AI as a helper running a parallel set of tasks. It’s the hardest step to take, because it changes who does the work and who’s accountable for it. It’s also where leaders set themselves apart on agents, from how widely they run agentic AI to how much room they give it to act.

They go wide and deep. Leaders have new AI broadly deployed or embedded in a little more than half of their business areas. The bottom third? 19%. Leaders aren’t running a pilot in every department and calling it a strategy. They’re picking what works and spreading it across the company in the business activities where what matters can be measured.

They aim AI at the work that touches revenue. The biggest gaps between leaders and laggards aren’t in the obvious places like customer service chatbots. About half of companies use them, and leaders and laggards are about equally likely to, 50% vs. 48%. They’re in the work that moves the needle. The activities that touch revenue and cash. Leaders are more than twice as likely to use AI for contract and proposal generation. They’re 72% more likely to use it for invoice processing and reconciliation. And 61% use AI for incident detection and response, compared with 32% of the bottom third.

Read More: AI Agents Could Fill Treasury’s Looming Talent Gap

Think about what those jobs have in common. Proposals bring money in. Invoices get it collected. Incident response keeps it from walking out the door. They’re also exactly the kind of multi-step, rules-heavy work agents are built for.

They spend like they mean it. Leaders spent an average of $14.5 million on new AI in the past year. The bottom third spent $9.4 million. That’s 55% more. And 22% of leaders plan to raise their AI budgets by more than 15% next year, compared with 8% of the bottom third.

They have conviction.  Leaders are more likely to call AI and agents a fundamental, game-changing shift for their industry. They’re not AI doomsayers. They’re far more likely to say AI’s benefits outweigh its ethical concerns, 63% vs. 38%. That’s a big turnaround. Back in 2024 and 2025, the AI leaders were actually less likely than others to say that. Today’s leaders have gotten close enough to the technology to trust it, to see its impact and to see where it still needs more hands on the wheel.

The Agentic Enterprise Gap Is Likely to Be Widening

If most companies sitting below the agent line makes you think there’s plenty of time, look at how fast those who’ve made progress have climbed so far.

In 2024, GenAI showed up as an all-purpose assistant. The average big company ran about 6 of the 13 GenAI uses we tracked, mostly routine work like drafting content and answering customer questions. That’s the first rung. Use.

By early 2025, GenAI had moved into daily work. Executives told us 46% of their GenAI tasks were running mostly without human intervention, and they had high or complete trust in the output for nearly all of them. That’s adoption, edging into embedment.

Then agents arrived, and in about a year they went from something few companies were testing to something nearly every company we surveyed was using somewhere.

The money has followed the trust. In early 2024, 20% of executives called their AI ROI very positive. In 2026, 57% did. That’s more than the share of 2024’s AI leaders who said the same thing back then. In two years, the average company caught up to where the leaders had been.

Read More: The Karen Webster Playbook for an Agentic World

That’s the pattern to pay attention to. Today’s leaders aren’t outliers. They’re a preview. If the typical company follows today’s top third the way it followed 2024’s leaders, half of its business areas would have AI broadly deployed or embedded by 2027 or 2028, up from about a third today. And the next climb will be across the agent line.

The companies that wait to see how this plays out will still be doing the equivalent of comparing airfares in six browser tabs.

The Hard Part of Enterprise AI Adoption Isn’t the Technology

So, here’s a question. If the companies furthest along are also growing fastest, why isn’t everyone putting agents to work?

Ask executives and they’ll tell you. Half of them say their biggest limit on AI isn’t the technology. It’s people, processes and culture.

That tracks with everything I hear in boardrooms and with the CEOs I talk to. The models are amazing and getting better every day. The tools are easy to use. What’s missing is the willingness to redesign the work around agents. And to trust them with decisions that matter.

Read More: Why Building AI Agents Is No Longer the Hardest Part of Agentic Commerce

Crossing the agent line isn’t a software upgrade. Below it, AI helps a person do the job. Above it, an agent does the job and a person approves it. That’s a change in who’s accountable, how the process flows, which roles exist and what people spend their days doing.

That’s hard. It means rewriting job descriptions, changing approval chains and deciding, ahead of time, where a human has to stay in the loop and where one doesn’t. It means having governance strong enough that the risk team says yes instead of no.

Read More: AI Has a Frankenstein Dilemma

It may also explain why security and data governance teams are moving first. Teams like these have typically spent years learning how to trust a model, check its work and widen its role as it proves itself. The rest of the business can learn a lot from how they’ve done it.

And there’s a workforce story here that’s more interesting than the one in the headlines. The bigger story is growth. Eighty-four percent of executives expect AI to let them grow without a matching increase in headcount, and 82% expect to raise productivity with headcount held steady. Three in four expect roles and skills to shift. The companies that get the people-side right will get to grow without hiring at the same pace. The ones that don’t will stay below the agent line, no matter how much they spend on tools. Or people.

How to Get the Agentic Edge

Think back to that travel agent. People trusted that person with their vacation because they’d earned it, one trip at a time. And it was more efficient than doing it themselves. Agents will earn trust the same way. Start with a person approving the work, and widen the agent’s job as it proves itself.

The money to do it is coming. Seventy-nine percent of big companies plan to raise their AI budgets next year. Not one of the companies we surveyed plans to cut. The question is where it lands. On yet another helper, or on agents doing the work that touches revenue.

So where should that money go? Here’s how to get the agentic edge, starting Monday morning.

1. Stop counting use cases. Ask every business leader where AI sits on the ladder for the work they own, and whether people or agents are doing it. Focus on the activities that really matter to revenue, margin and risk.
2. Start where your AI is oldest and most proven. In most companies, fraud, security and risk teams have worked with machine learning for years, and security monitoring and data governance are already where companies are most willing to let agents act. Make them the proving ground, with governance to match.
3. Aim agents at the work that touches revenue. Proposals, invoices, incident response. The places where leaders pull furthest ahead are the plumbing of the business, not the showroom. And where it is easy to measure return on investment and business impact.
4. Pick one process and let an agent run it. Choose routine, high-volume work, give an agent the job, keep a person approving it and widen the agent’s role as trust builds. That’s crossing the agent line, and it’s where leaders set themselves apart on agents.
5. Budget for people, not just licenses. If half your peers say people and process are the bottleneck, they probably are for you, too. Fund the redesign, the training and the new roles.

Then, in six months, ask the question I posed at the beginning. Who’s doing the work?

If AI is still making suggestions while your people do the job, you’re using AI.

If agents are doing the work while your people check it, congratulations. You’ve got the agentic edge. And based on everything we’ve seen over the past two years, that’s where the companies we see growing faster today are heading, and the kind of company we expect to still be ahead when the rest of the market finally starts asking the right question.

Until NEXT time.

Join the 21,000 subscribers who’ve already said yes to what’s NEXT.