Exclusive: 34 CEOs on what thrills and terrifies them about agentic AI 

When businesses leaders think about AI right now, they’re thinking about how agentic tools will change the very nature of their work. “We’ve crossed a line,” says Varun Krishna, CEO of the fintech giant Rocket Companies. “AI is no longer just creating. It is thinking, deciding and acting. That changes everything, from client interaction to security.” As agentic tools get more sophisticated, they “expose how many organizations are still constrained by siloed data, fragmented ownership, and legacy ways of working,” says Ndidi Oteh, who leads Accenture Song, the professional services and creativity arm of the IT company Accenture. “Companies that don’t rewire how work gets done will slow themselves down at exactly the moment speed matters most.” Even those skeptical of how real agentic tools are recognize that they can wholly transform business. “Most of what people are calling agentic is just automation with better marketing,” says Meng Ru Kuok, CEO of Caldecott Music Group, owner of the AI-powered music production tool BandLab. “The genuinely autonomous systems—ones that negotiate, transact, and make judgment calls without a human in the loop—those aren’t here yet for most businesses. That said, the potential for agentic AI to empower people is absolutely real.” This spring, Fast Company asked almost three dozen CEOs—in industries ranging from healthcare to marketing and including the leaders of Abridge, American Express, Anduril, Sephora, and the ad agency Johannes Leonardo—how they’re thinking about agentic AI: their strategies, the tools they’re building, what they’re excited about, and what scares them. Among the clearest themes that emerged from CEOs’ responses: Nobody is ready for how agentic AI is going to transform business, because companies are still figuring that out. But as uncertain as the agentic future’s exact contours are, it’s equally clear that companies should already be sketching them out. As Charles Yang, cofounder and CEO of the AI workspace company Vibe, puts it, “I don’t think readiness is the right frame. The right frame is: Are you in the water? Because you can’t learn to swim from the shore.” Agentic AI will reshape productivity Agentic AI may be growing, but it’s still in its early stages. Despite that, executives are already extrapolating productivity gains that agentic tools might bring based on how AI tools are already changing the way their organizations work. Steve Squeri, CEO of American Express, notes that with AI-assisted development tools, the company has scaled them to the point that coding cycle times are down 30, speeding up the software development process. “Agentic AI takes that a step further,” Squeri says, “moving from tools that support developers to systems that can increasingly take on full ownership of designing, testing, and deploying new capabilities. Over time, that has the potential to fundamentally change the speed and scale at which we innovate.” Squeri says he can see a future where AI agents handle entire trips for cardholders—booking airfare and lodging, and even creating a personalized itinerary with reservations through Amex’s subsidiary Resy. “If we do this right, the best agentic AI experiences won’t feel like AI at all,” he says. Customer-facing autonomous tools aren’t entirely conjecture—at least not for Arintra, an Austin-based AI medical coding company. Its CEO, Nitesh Shroff, says the company has built an agentic AI coding engine that sits within a provider’s electronic health record that can turn clinical documentation into billing codes using insurance-specific rules. At the health system Mercy Health, Shroff says Arintra’s tools autonomously process 50,000-plus charts a month, helping boost Mercy’s revenue 5.1. “What we’ve seen is that agentic AI doesn’t replace the expertise of human coders,” he says. “It gives them the space to do the more complex work that they were trained for, instead of the routine high-volume work they can’t keep up with.” Efficiency is great, but there are downsides Helping teams of any size work faster is a big advantage that companies see as they explore agentic tools. But the advantages can be quickly undone. On the one hand, there’s what Nick Deveau, cofounder and CEO of the AI-based leasing tool company Grotto AI, calls “the mirage of productivity” being driven by AI. “It’s easier than ever for me—within the span of 10 minutes—to work on optimizing my Google ads campaign, dive into data analysis, submit some new code to our codebase, and spitball a new strategy. You feel like you’re on fire,” he says. “But is the output quality good? Can you truly produce high-quality results with so much competition for your focus? We can move extremely quickly now, but it’s easier than ever to move in the wrong direction even faster.” Beyond quality of work accompanying increased efficiency, there’s also the fundamental challenge of autonomous tools acting on their own. “When an agent makes a bad procurement decision or a flawed design call, who is responsible?” says Nathan Silvernail, CEO of Plantd, which builds structural building panels out of grass rather than wood. “Most companies don’t have answers to that yet.” One of the biggest areas of agentic potential is in marketing, where executives are excited about the speed that it can bring to organizations of all sizes. Helen Andrews, CEO of the advertising agency Johannes Leonardo, says agentic tools mark a big shift from other tech that has become standard across industries. “The past 20 years of technology mostly added work,” she says. “More platforms, more formats, more reporting, more coordination. Most of it shallow. Agentic AI is the first thing in a long time that could actually subtract from that pile. Hand the false busyness to agents. Give humans back the hours that produce the real stuff: the unexpected insight, the relationship, the idea that no one saw coming. Creativity.” Even as agentic AI frees up creatives to do more visionary work, Jason Harris, cofounder and CEO of the creative agency Mekanism, says that the discourse around agentic AI and its ability to replace workers is a big concern, not because it will replace workers, but because “clients convince themselves it already has.” The result, he says, is “the commoditization of ‘good enough.’ Agentic AI can spin up briefs, generate concepts and optimize media buys without a single moment of instinct or risk-taking. A procurement team looking at a spreadsheet can’t tell the difference between work that’s efficient and work that’s alive.” It’s sentiment that’s shared by another agency executive Sean McDonald, Rethink’s global chief strategy officer and partner. “My biggest concern with regards to AI is where it might lower the bar and tempt marketers with ease and speed over quality and impact,” he says. “We have seen a litany of technology-driven hype cycles come and go, promising significant cost and efficiency gains while glossing over the negative impact on quality.” John Elder, CEO of the Supergood agency, feels similarly. “The real risk is that every brand gets access to the same off-the-shelf agents and the work becomes indistinguishable,” he says. “If everyone’s prompting the same models with the same data, we all produce the same mediocre middle. That’s why we invest 74 of our stack in proprietary data. AI will be evenly distributed. Data won’t.” Speaking of data . . . Data will be your advantage—so will good governance “There is a data quality problem sitting underneath all of this,” says Taryn Crouthers, CEO of the digital media firm Big Spaceship. “Agents are only as good as what they’re fed, and most organizations don’t have the clean, structured, accessible data that agentic systems require. This is going to be a silent bottleneck for a lot of brands. Some organizations are deploying agentic AI on top of org charts and data infrastructure that were never designed for it.” In industries like healthcare, where Abridge deploys its AI clinical documentation tools, data will be the differentiator, but it’s up to systems to make sure data can travel between places where it’s needed. “Agentic systems are only as powerful as the data they can access and act on. If clinical, financial, and administrative data remain siloed, fragmented, or delayed, it constrains what agents can actually do—no matter how advanced the models are,” says Shiv Rao, CEO and cofounder of Abridge. “At the same time, if data does become highly portable and standardized, it raises a different risk: Value shifts quickly to whoever can orchestrate that data into action in the safe and reliable way that healthcare requires.” Rao’s assessment might be optimistic. Ali Diab, cofounder and CEO of the employer health plan administration company Collective Health, doesn’t see much of healthcare adopting agentic tools anytime soon. “We are still a country where fax machines are a primary mode of communication and ‘proprietary data’ is used as a weapon to prevent competition,” he says. “Agentic AI requires high-fidelity, real-time data to function safely. If the underlying data is garbage—or worse, intentionally obscured—the AI will fail.” Many of the same challenges that plague healthcare span industries. Mariam Hakobyan, CEO of Softr, which enables businesses to create enterprise apps with no code using AI, says the groundwork of data and governance is still being laid. “Readiness isn’t just about willingness. It comes down to infrastructure, trust, and understanding,” she says. “Most companies still lack clean data, well-defined workflows, and strong governance. The shift is clearly coming, but for many traditional industries, adoption will take time.” Better data isn’t a magic bullet, though. According to the CEO of a leading marketing and advertising firm, who opted for the anonymity we offered to speak candidly, thinking that having data is sufficient to change how things are done is giving up the game before it starts. “[Companies are] still shaped by the Old World-notion of a company provides a product or service and the consumer buys it. Then they say AI and data will help that become more personal. That is a very narrow lens,” the CEO says. “Better data doesn’t just mean that I know now if you’re into this or not. You are going to have an ongoing relationship with your customers, and then you have to think about what that means.” Every agent can be a security vulnerability—or a hacker Healthcare isn’t the only industry thinking about how data can be secured. Agentic AI brings with it various security threats, both internal and external. It was the biggest concern identified by many CEOs surveyed, and Logan Brown, the Harvard Law-educated founder and CEO of the legal AI company Soxton, puts it plainly: “The increase in capabilities of AI is going to lead to an unprecedented level of malware, hacks, and security issues. I am concerned about the risk that agents interacting on the web will pose to business and humans.” That fear exists across industries. John Levy, who leads the quantum computing company SeeQC, says security needs to be on everyone’s mind. “Most organizations are still figuring out how to deploy even basic AI tools safely. Agentic systems, which can take autonomous action, raise the stakes significantly,” he says. “A significant portion of our value resides in unpublished research and prepatent innovation. Agentic AI can accelerate workflows like prior art searches and patent drafting. But it also creates real risk if sensitive information is exposed through external models or insecure environments.” Agents don’t just create risk when deployed by businesses. Their ease of use also makes them easily enlisted by malicious actors to supercharge scams and malware. “The latest AI models give both defenders and attackers access to powerful tools,” says Navrina Singh, CEO of the AI governance company Credo AI. “That creates a new class of ‘unknown unknowns.’ We don’t yet fully understand the attack surface these systems introduce, and that uncertainty is deeply concerning.” Ben Colman, who leads the deepfake detection company Reality Defender, actually knows a lot of those unknowns. His company knows that AI agents are helping scammers—even small operations—to the next level. “Bad actors are using AI voice agents at scale to make calls, start social engineering campaigns, and generally wreak havoc: Think 10,000 calls being made to a call center to waste their resources or seek the PII [personally identifiable information] of a person (or people) who have an account with a specific company,” Colman says. “As the least secure person is a company’s greatest security vulnerability, it just takes one of those calls to part with information it should not have for such attacks to be a success.” He adds that the ability for agents to run cheaply in parallel, and generate synthetic voices in real time, means that “it’s impossible for the average or even the most astute user to determine whether or not they’re speaking with a real person or an AI agent, which is something that our clients and industry peers are concerned about.” Autonomous tools still need people at the center Whether internal or customer-facing, agentic tools are still tools. CEOs who participated in this survey largely recognize that, regardless of how this technology fits into their business. From the consumer side, Sephora‘s global president and CEO Guillaume Motte says that the retailer’s efforts to deploy agentic tools to undergird customer experiences isn’t coming at the expense of shopping’s interpersonal elements. “Beauty is built on trust and connection. Technology must strengthen and complement—not replace—those fundamentals,” Motte says. “For Sephora, our interest is to use it well to enhance the consumer journey, like navigating complexity, understanding products, and making better-informed choices. At the same time, beauty will remain a client-focused business and our beauty advisors will always be essential to interpretation, empathy, and reassurance.” Maintaining the human element amid increased use of AI in the creative sector is also critical, says Jennifer Gonzalez, CEO of the Studio Marie agency. “Human friction and struggle are genuinely productive for us,” she says. “Chance interactions during our days often yield as much as a landscape deep dive, design ‘mistakes’ become new ways into concepts, and rethinking templates keeps us challenging our own assumptions. Our edge has always been an instinct for pairing the right creative voice with the right problem and that’s something tools can’t replicate.” Your workers can probably stick around—but might need some coaching It is likely true, as Ramy Khorshed, CEO of the Egyptian property tech company Sakneen says, that “employees face more threat than businesses; the majority of businesses will find it much easier and cheaper to implement major operational efficiencies due to agentic AI.” But responses from a large swath of business leaders we surveyed show that many executives are more circumspect than simply equating efficiency with an opportunity to cut headcount. At best, Jarek Kutylowski, CEO of the AI translation company DeepL, sees an autonomous future of workers whose job is managing outcomes rather than solving problems. “In many ways, it is forcing every employee to stop being a builder and start being a CEO,” he says. “A CEO’s role is about overseeing a system that is far too vast to probe in every minute detail. Instead, you have to decide what to trust versus where to manage risk. With agentic, we’re essentially asking people to move from knowing it all, to the discipline of high-level oversight. This is a massive opportunity for the workforce to prove their judgment in the same way a leader does. Helping people redefine their professional expertise and interests to account for this shift is also a challenge.” Robin Forbes, CEO of the creative agency R/GA, sees the opportunity for workers—especially creatives—to change how they view their jobs and how they end up billing for work. “With agentic AI enabling creative service businesses to work more efficiently, time can no longer be viewed as the main unit of value. Value needs to be measured by the quality of the output and impact on business outcomes,” he says. “By anchoring success in quality and business outcomes rather than hours, we also validate the strategic value of our talent. This ensures our people are recognized as the architects of impact rather than being relegated to machine operators in the name of efficiency.” Janet Sherlock, CEO of the organization design firm Org.works, says that a transformation of workers’ roles needs to be accompanied by concurrent changes in how leaders define roles. “Most companies haven’t answered the basics: who decides where agents are used, who owns the outcomes, and what platforms they’re standardized on,” she says. “When decision rights and accountability aren’t clear, agentic AI doesn’t create efficiency; it amplifies confusion.” When thinking about challenges as companies experiment with agentic processes, two CEOs used calculators as an example of what widespread agentic tools could do to the workforce. Stephen Smith, founder and CEO of the mental health platform NOCD and its parent software company Noto, says: “The biggest risk I see is that our team will over-leverage AI to the point where their foundational skills as operators erode. Reflect on using a calculator. Do you still know how to do long division, or have you forgotten due to your reliance on using a calculator? If you overutilize AI for problem solving and writing, what will happen to your core problem solving and writing skills? They might erode, unless we intentionally keep growing.” The need for continued growth is an area where Bijal Shah, CEO of the workforce education benefits company Guild, sees an opportunity for companies to invest in upskilling their employees. “The businesses that are preparing [for the agentic future] are not necessarily the ones with the most advanced technology. They are the ones creating the conditions for learning, giving people time to experiment, sharing use cases, and normalizing the friction that comes with learning something new,” she says. “Every major technology shift has followed this pattern. The difference now is the pace. Therefore, being ready means investing in people as intentionally as we invest in technology.” The potential for investment in agentic AI to go hand-in-hand with workforce transformation is a prospect fueling the defense tech startup Anduril, whose CEO, Brian Schimpf, views agentic tools as a path to a resurgence in U.S. economic strength in making physical products, not just knowledge work. “The singularly most underdiscussed part of the AI transformation is how we can use the technology to increase industrial resilience and expansion—not threaten the future of work,” Schimpf says. “So far, early economic reporting and predictive analysis alike predict steady and ready growth in the job market. Especially in physical industries, agentic AI is likely to be additive to output and increase demand, rather than to shrink either. At Anduril, we view the technology as a way to rebuild military power and bolster the U.S. industrial base at the same time, rather than subscribing to the view that AI is a harbinger of economic mass extinction.” Nobody feels ready. Start anyway Our survey concluded with the question: “Are businesses and consumers ready for the sort of sweeping change agentic AI may bring? Why or why not?” Nobody responded with an unqualified yes. Most said no. Some said consumers are further along than businesses. Part of that is the sheer speed at which AI and agentic tools have proliferated. “Never before have so many fundamentals been in flux simultaneously: models, agents, interfaces, distribution, measurement, and the economics of the open web,” says Judith Carr-Rodriguez, partner and CEO at the independent creative agency Fig. “Previous shifts gave you time. Broadcast had decades before cable fragmented it. Digital had years before mobile reshaped it. You could adapt. What’s different now isn’t that change is happening; it’s the number of things changing at once and the speed each one is moving.” But respondents also agreed that they should start identifying the competitive advantages of agentic tools as soon as possible. Vikram Bhaskaran, CEO of the physician-only knowledge network Roon, says the way consumers interacted with AI in healthcare is instructive of how business can take the tools that exist and just start experimenting. “For years, patients turned to ‘Dr. Google’ to fill that gap. Now many are turning to tools like ChatGPT for faster, more contextual answers,” he says. “That shift isn’t happening because people think AI is perfect; it’s happening because access to human expertise is constrained. Agentic AI is just accelerating a behavior that’s already underway.” And though first movers may be at an advantage, another anonymous marketing and advertising exec advises business leaders that with a technology as nascent as agentic AI, they should be gearing up for a marathon, not a sprint. “This might be nerdy, but I think the German sociologist Niklas Luhmann is instructive here: People tend to underestimate system inertia, i.e. people are not ready and they will slow down the development. That’s good and healthy,” this exec says. “Not everything that is possible today will happen at scale tomorrow. It’s going to be a long and winding road.”