Companies cannot workshop their way into understanding AI. At some point, the technology moves beyond the pilot and into real daily workflows. Over the course of two days at Reuters Momentum AI, case studies, panel discussions, and executive presentations all spoke to the same thing, just with different contexts. The question the room kept returning to: now that AI is no longer optional, how do we actually run it?
The honest answer, across nearly every panel, was that no one has fully figured it out yet. But the conversation has clearly moved. It is no longer about who is using AI—it is about how, with what guardrails, and at whose risk. The word that surfaced again and again, on stage and in the hallways, was governance: a catch-all for the integrity, transparency, and accountability the next phase of AI is going to demand of everyone deploying it.
One shift surprised me. Since the beginning of AI’s rapid expansion, employees treated the tools as something to hide—used experimentally on the side, kept out of view for fear of seeming lazy, unethical, or like they were cheating somehow. That has flipped. The strongest message across all panels was that AI isn’t a tool for replacing human work, but one organizations need to be open about, so teams can adapt, share what works, and learn from each other.
The stage reflected the breadth of what that actually looks like in practice. The audience erupted as Aaron Levie, co-founder and CEO of Box, took the stage to announce Box Automate, which lets AI agents handle tedious tasks like processing invoices. Luke Gebb, Global Head of Innovation at American Express; Fiona Tan, Chief Technology Officer at Wayfair; and Tom Ellis, Head of Consumer and Technology and Chief Information Officer at Bank of America were among those walking through how their teams are integrating agents into real work. Sherry Sanger, EVP Strategy and Marketing at Penske, shared how the company is using AI to track its trucks across the country and save customers both time and money. Campbell Brown, co-founder and CEO of Forum AI, took the risk side, focusing on what we don’t yet ‘know.
Surrounded by about 100 technology experts, entrepreneurs, and business leaders, I was surprised by how welcoming the room felt—considering I am none of those things. The atmosphere was more curious than pretentious. Everyone, regardless of seniority, seemed to recognize they were running an experiment together. We were all engineers in this grand experiment.
Five themes carried across the two days.
Leadership in the Age of AI
Michael Domanic, Head of AI at Section AI, wants everyone to stop saying: “AI won’t replace you, but someone using AI will.”
It’s a phrase that appears everywhere—from LinkedIn posts to conference keynotes—and according to Domanic, it’s doing real damage. “That line wasn’t even true three years ago,” he told the audience. “And it’s even less true today.” Employees hear it and feel dismissed almost immediately. The more it circulates, the more it erodes trust in the leaders saying it. His replacement framing: “Work is changing, not disappearing.”
That distinction matters. The gap between what executives believe is happening with AI adoption and what employees are actually experiencing is staggering. Domanic cited a survey of 5,000 knowledge workers: 57% of C-suite leaders say their organizations have widespread AI adoption, but only 26% of the managers below them say the same. From the same survey, 76% of executives believe employees feel encouraged to build their own AI solutions, when in reality, only 24% of employees feel that way. “That’s not just a gap,” Domanic said. “That’s a completely different reality.”
He grouped these employees into four main pools. The first: the 10–15% of employees who are active resisters. These are the conscientious objectors who have principled or personal reasons for pushing back against AI adoption. The second being the 30% who are simply disengaged, indifferent—or waiting to be convinced. Then we have the 30% who are genuinely excited but have no idea where to start. And finally, you have the 10–15% who are power users. These are the folks who have fully adopted the technology and woven it into their day-to-day workflows. Most of whom often feel a bit of frustration—either that the tools aren’t as good as they’d hope, or that the rest of the organization isn’t heading full speed towards adoption alongside them.
Most AI adoption strategies are aimed at the first three groups. Only a small number are designed for the fourth. This is where a lot of institutional momentum is being left untapped.
The harder truth is that executives don’t just face an implementation problem—the technology itself is relatively accessible and affordable. The challenge is getting employees to not only integrate it, but also adapt to it. Domanic’s session was built around the five questions that are actually circulating in team meetings, one-on-ones, and Slack channels.
“Isn’t AI killing the planet?” This is where empathy is crucial. By grounding your response in real data, clearly stating your organization’s direction, and not trying to cover up the cost on climate, you empathize with them, without sweeping their concerns under the rug.
“Am I training my replacement?” The job isn’t going away, the composition of it is.
“Isn’t this all AI slop?” Employ the driver vs. passenger philosophy to this question.
“What’s our competitive advantage if everyone has the same AI?” Remember that in a lot of cases, AI is leveling the playing field. If used effectively, the growth potential is enormous.
“Is this even worth it?” Remind them that most experiments fail (roughly half). Encouraging them to try and to fail and finally start again using cases of their own and others is the fastest way to see success.
The State of Agentic Commerce
One industry that’s taken AI and run full speed ahead has been commerce. There’s no denying that traditional brick-and-mortar shopping has been declining for years, but without schlepping it to a half-abandoned mall 30 minutes away, shopping is still a strenuous process full of unwanted clicks and sifting through 13 toilet paper brands to find the cheapest option.
The real progress showed up in deployment. These were the brands that jumped first and looked second.
Target: Target’s SVP of technology, Brad Thompson, has been working with Target’s team to streamline this process. Target became the first major retailer to integrate with ChatGPT—and built the integration in three weeks… during Black Friday. Target’s use of AI has made its way to Microsoft Copilot, which lives within the native checkout. The example Thompson showed was planning for a movie night. The agentic commerce technology can curate the perfect movie night with a simple prompt, healthy snacks, cozy blankets, fuzzy socks, and a sweet treat. Without even leaving the app, your virtual shopping cart has curated the perfect movie night, within your budget, without any added hassle of remembering that popcorn brand your kids loved.
Wayfair: Wayfair is taking a similar approach to agentic commerce, with their dual strategy. This dual approach is a cohesive way to shop both on the site, and via an external AI platform. Wayfair’s CTO, Fiona Tan, shed light on how the company is working on Model Context Protocol (MCP)-based agent capabilities on their own platform. The MCP is a dynamic shift in how AI agents actually interact with a product, and services. The MCP then sits in front of all the other Application Programming Interfaces (API), ensuring that the agent’s capabilities are built upon actual knowledge of the company’s services. Once the MCP server is built, other agents will all gain full access, for free. Therefore, compounding returns overtime.
AMEX: AMEX is also throwing its hat in the ring of Agentic Commerce by developing their trust frameworks. A few examples include: KYA, Verified Intent, and Agent Purchase Protection, all working to solve the accountability gap that AI agents can create. Agents are a third entity that will make errors—but unlike merchants or cardholders, they won’t answer the phone. To combat this, AMEX they have deployed their agent protection program—take on the risk if intent is verified and data is shared.
The panel had one major takeaway, e-commerce didn’t kill stores, and mobile didn’t kill websites—all channels survive but find their level.
AI Infrastructure
GPUs, training clusters, and frontier models have dominated the AI infrastructure story. But as enterprises move from pilots to production, the bottleneck shifts to something less glamorous: how fast and efficiently data can be stored, moved, and accessed.
Greg Matson, Head of Marketing, Business and Product Strategy at Solidigm, sat down with Jeff Denworth, co-founder of VAST Data, to explain. Their core point: inference at scale is fundamentally a data problem, not a compute problem. A bad storage strategy reduces the ROI of every other infrastructure and software investment a company makes. GPU dollars carry only as much weight as the data feeding them.
Customer Experience
Puneet Mehta, founder and CEO of Netomi, was direct: enterprise AI desperately needs a win.
The current landscape can only sustain unproven ROI for so long. The three obstacles, in his view, are predictability, speed of iteration, and a measurement framework that hasn’t caught up with how fast the technology is changing. He proposed retiring the CapEx + OpEx model that has dominated tech budgeting for the last decade. Projects launched six months ago are already outdated when a new model arrives. The rapid evolution of agents has improved workflows, but it has also produced a kind of whiplash. Cost savings get the headlines, Mehta argued, but the real upside of AI-powered customer experience lies elsewhere—in customer lifetime value, larger transaction sizes, upsell opportunities, and the harder-to-quantify but very real metric of customer happiness.
Risk and Security
Risk and security has been a long winded slap on the wrist when deploying AI agents into workflows. The insurance industry has been among the first to feel the true threat of trust. Take the world’s largest insurers most of which are now choosing to exclude AI-related damages from standard policies because there is no baseline to measure and no standard actuarial model exists for probabilistic systems.
A useful frame came up more than once: agents are interns. The message here being that you wouldn’t trust an intern to handle client sensitive information without looking over their shoulder, so why would you allow your AI agent to do the same? Something that goes without saying, depending on the future of AI, is risk, and it will stay top of mind.
Momentum AI answered a lot of questions. AI is no longer something happening at the edges of business, it is becoming embedded in the core of how decisions are made, how trust is built, and how value is created. But one still remains, who will be the winners that adopt it thoughtfully, with the governance, transparency, and leadership this next phase demands. That conversation is far from over. It continues at Techonomy 26, where the focus turns to what intelligence at scale actually requires from the people building and deploying it, and what it will take to get it right.
