Key takeaways

• Physical AI is fundamentally harder because “the variables are very different” and real-world complexity is less predictable than in purely digital systems
• Enterprise value comes from prioritization and discipline: “I don't ever want to automate a bad process”
• Trust must be engineered through customer choice, not imposed: “choice in the hands of our customer”
• Skilling is the biggest readiness gap, with renewed importance of the “humanities muscle”

• AI success depends on change, data and action: “how you manage the change matters,” “data quality,” and “bias for action”

  Artificial intelligence has moved well beyond experimentation, but its next phase is being defined by something far more impactful than incremental automation. As AI systems begin to reason, adapt and operate with greater autonomy, organizations are being pushed to rethink how value is created, how work gets done and how humans and machines collaborate at scale. This shift marks a move from isolated pilots to AI-first strategies that span physical and digital environments and the workforce itself.

  This reality framed a wide-ranging conversation in the Ice Village at the World Economic Forum, where senior leaders from across industries came together to examine how AI is translating from promise to success. Moderated by Louise Victoria Matsakis, Senior Business Editor at Wired, the panel brought together Vijay Guntur, CTO and Head of Ecosystems at HCLTech, Sheila Jordan, Senior Vice President, Chief Digital Technology Officer at Honeywell, Aparna Bawa, Chief Operating Officer at Zoom and Dr. Guy Diedrich, Global Innovation Officer at Cisco to explore what it truly takes to deploy AI responsibly, securely and at scale.

  Rather than focusing on distant possibilities, the discussion centered on what is happening now: how advanced AI is reshaping enterprises, consumers, factories and critical infrastructure, embedding itself into everyday collaboration, redefining workforce skills and forcing enterprises to confront hard questions around trust, governance and change. Across every perspective, one message was clear. The organizations that succeed in this next era of AI will be those that pair technological capability with disciplined execution, human judgment and the courage to move from experimentation to action.

  Designing AI for the physical world

  AI behaves very differently once it leaves the digital domain. As Guntur explained, “The physical world is very different. The variables are very different. Physics needs to work.” Unlike enterprise software, Physical AI must operate under real-world uncertainty, where “the number of combinations and the amount of data that you need to deal with is enormous.”

  This complexity is why simulation has become central to deploying AI safely and effectively. “The core to this is about simulations,” said Guntur, describing how teams model thousands of scenarios before systems ever touch a factory floor. But simulation is only the beginning. Reality will always introduce surprises. “In real time, something else crops up,” particularly in domains like autonomous driving, where unexpected edge cases must be captured, analyzed and folded back into training cycles.

  Despite rapid advances in compute and edge infrastructure, Guntur stressed that autonomy does not remove human responsibility. “You need to still have...a human in the loop,” especially when systems control fleets, factories or other critical assets. The role of humans shifts from execution to oversight, judgment and intervention.

  Turning pilots into measurable business impact

  For Jordan, AI success starts with focus. At Honeywell, the strategy is to deploy only what materially moves the business. “We’re taking an approach to really prioritize those things that have a high impact, either value from productivity or revenue,” she said.

  She explained that technology maturity is no longer the limiting factor. “The technology is here,” she noted. The real work lies in operational readiness. “I don’t ever want to automate a bad process,” she said, warning that AI amplifies inefficiencies as easily as it improves outcomes. That is why process simplification and data quality must come first. “You can’t have a GenAI or agentic strategy if you don’t have a data strategy. Data is so important.”

  Crucially, Jordan emphasized that AI’s biggest returns often come from crossing organizational boundaries. “The value occurs when you actually deliver an end-to-end process for your customers and partners,” she said. With agents now capable of orchestrating workflows horizontally, AI can connect marketing, sales, finance and operations in ways that traditional automation never could.

  Building end-to-end AI experiences

  That horizontal value is illustrated by Honeywell’s “superstar seller” concept. Rather than supporting sales through fragmented tools, AI can guide the entire journey. “We can actually provide all the content about the customer,” explained Jordan, “have the agent help us build the strategic account plan,” and then recommend “products and services” based on existing deployments. “So, it’s that end-to-end,” she said, where context, planning and execution converge.

  Bawa described a similar philosophy at Zoom, grounded in a clear principle: “Our vision is to make AI enhance human connection, not replace it.” By embedding AI across the platform, Zoom aims to eliminate routine work and enable better collaboration. “It will create action items for you after the meeting…It might take notes for you,” she said, turning conversations into outcomes.

  Unexpected adoption patterns further reinforce this value. Bawa pointed to Zoom’s webinar functionality as an example, describing how it is used across a wide range of contexts, from political campaigns to large-scale marketing engagements. Rather than being defined by a single use case, the platform’s impact lies in its accessibility at scale. As Bawa framed it, “It’s democratizing participation.”

  Trust, privacy and speed by design

  At scale, trust becomes an architectural requirement. Zoom serves customers ranging from individuals to global banks and “everybody has a different set of risk tolerance,” said Bawa. The solution is not a one-size-fits-all policy, but flexibility. “We put choice in the hands of our customer so they can make the trade-off between speed of innovation and security and privacy.”

  That flexibility allows one customer to mandate transcription for compliance, while another can insist that no transcription take place at all. Trust is reinforced through explicit boundaries, including Zoom’s commitment not to use customer data to train its models. That stance reflects Zoom’s position at the center of how work happens, where conversations translate directly into decisions and actions.

  Skilling and the return of human judgment

  Diedrich placed enterprise challenges in a global context. Cisco’s Country Digital Acceleration program emerged because nations invested in digital strategies but “they got a PowerPoint” from big consultancies. Cisco stepped in as a partner to help build an execution plan for a national digital architecture. Then, co-invest and co-develop individual projects to show the value and impact of digitization. Today, priorities have shifted. “Now it’s all about skilling,” he said.

  Cisco’s data underscores the urgency. “[Only] 13% of companies consider themselves ready for AI,” said Diedrich, citing Cisco’s 2025 AI Readiness Index. While infrastructure and security matter, digital literacy is the foundation.

  What’s missing is not technical expertise. “We have incredible technical muscle,” he said, “but we also need to have cognitive muscle.” As AI becomes ubiquitous, the differentiator becomes judgment. “What’s going to become the most important skill moving forward? It’s asking the right questions.” He argued for integrating ethics, problem-solving, psychology and sociology into technical education to create more resilient leaders and teams.

  Guntur echoed this, highlighting the growing value of “collaboration, communications and empathy” as coding and routine tasks become automated.

  Avoiding wasted spend and accelerating scale

  Asked where AI investments fail, Jordan pointed to prolonged experimentation. “POCs last too long,” she said and in today’s environment, “it’s more important to get progress than perfection.” Once value is clear, “It’s time to go to scale.”

  Diedrich warned against aspiration without execution. “Right now, we're seeing the resurgence of on-prem because of data sovereignty. I think you're going to see the same thing with AI. Right now, people are so aspirational and we need to ground ourselves in execution,” he said. The workforce implications heighten the urgency. “90% of the most in demand jobs in our industry are going to be dramatically impacted by AI over the next two years,” stated Diedrich, referencing Cisco’s AI Workforce Consortium reporting and 2024 projection that 92% of ICT roles analyzed would undergo moderate to high transformation. These roles will not disappear, but “dramatically change,” shifting skill requirements “to the right,” meaning that workers will be expected to operate at a higher level of strategic thinking and complex decision-making.

  Bawa reinforced the need for pragmatism at scale. Enterprise AI adoption, she noted, can’t be treated as “one big monolith stuck in time.” Risk appetite and readiness vary widely across customers and are evolving quickly. Platforms that succeed are those that give customers flexibility to move forward at the right pace, rather than forcing uniform deployment models.

  IT and OT convergence as the next frontier

  Jordan highlighted another inflection point: “the next big thing…is really being able to converge IT and OT.” Operational technologies often rely on “old legacy” systems designed to be isolated, yet AI-driven use cases demand connectivity and governance.

  That convergence is now possible, she argued, because “You’ve got 5G at the edge, you’ve got cloud compute and you’ve now got GenAI.”

  Guntur extended this thinking to critical infrastructure such as airports and ports, which already operate highly secure environments with extensive camera networks and sensor-rich operational technology. When that data is brought together, he noted, it enables entirely new experiences while improving safety and reducing incidents. He pointed to work underway with Los Angeles World Airports, where planners are looking ahead to major global events and asking how technology can improve the experience from the moment travellers land to the moment they leave. In this context, AI is accelerating a convergence that was already underway.

  Diedrich added a clear signal of momentum: “We are getting fantastic demand and feedback for a brand-new course called IT for OT, OT, [which] is being consumed all over the world.” Since 1997, the Cisco Networking Academy has provided free IT education to over 28 million learners across 195 countries.

  Determining AI success

  The panel closed with a succinct summary of what ultimately determines AI success. Guntur emphasized transformation: “how you manage the change matters.” Jordan named the foundation: “Data quality.” Diedrich stressed momentum: “bias for action. Move on it.” And Bawa returned to the human core, noting that success depends on employees being “open to always learning…not being afraid of change and embracing the change.”

  Together, these insights point to a clear mandate. AI success is no longer about experimentation. It is about disciplined execution, trusted systems, converged infrastructure and a workforce equipped not just with technical skills, but with judgment, empathy and the confidence to lead in an AI-powered world.

  FAQs

  What makes physical AI more complex than software AI? 
  Physical environments introduce physics, unpredictability and massive data variation, making deployment “orders magnitude higher” in complexity than digital-only systems.

  How can enterprises move beyond AI pilots? 
  By prioritizing high-impact use cases, fixing processes first and scaling quickly once value is proven rather than letting POCs run indefinitely.

  How does AI improve collaboration without adding noise? 
  By removing routine tasks, enabling translation and summaries and improving follow-through, AI can “enhance human connection, not replace it.”

  What is the biggest AI readiness gap today? 
  Skilling. Only “13%” feel ready and organizations increasingly need “humanities muscle” such as judgment, ethics and problem framing.

  Will AI replace jobs or tasks? 
  Mostly tasks. Jobs will “dramatically change,” shifting skill requirements higher and making reskilling a smarter strategy than replacement.