What the AI Strategy Summit Got Right (And What It Means for Enterprise Teams)
Earlier this month, a group of executives from NBCUniversal, PwC, Centene, Lumen, and Tech Mahindra gathered at the AI Strategy Summit from Section AI to talk about where enterprise AI actually stands. Not the pitch deck version. The real version.
A few things came up repeatedly across sessions. And the through-line isn't what most people expect when they picture an AI conference.
The conversation wasn't about models or tools. It was about people, process, and governance. The organizations doing this well have figured out that the hard part isn't the technology. It's everything around it.
Here's what stood out.
Section's biannual AI proficiency survey of 5,000 knowledge workers surfaced a number that should give every executive pause: two-thirds of employees say their organization has an AI strategy. But less than 20% say deployment is actually going well on the ground.
That gap is enormous. And it shows up in a few specific ways.
47% of employees say either no sanctioned tools exist or they don't know how to access them. Half the workforce says AI policy is unclear. 54% can't name a head of AI at their organization. And 38% have received no AI training from their employer at all.
The C-suite isn't seeing this clearly. 65% of executives say employees are generally positive about AI. Only 33% of individual contributors say the same.
This isn't a technology problem. It's a change management problem. And it's not getting solved by purchasing another platform.
La Sharell Morgan from NBCUniversal offered one of the clearest framings of the day. Governance, she said, is the guardrail on the side of the hill that lets people go up without falling off. Without it, you get two failure modes: people too afraid to use AI at all, and people going rogue with shadow AI.
Her practical approach: intake tracking is foundational. You can't govern, measure, or improve what you're not tracking. And not everything needs deep review. Out of 100 use cases, maybe 10 are truly high-risk. By defining what you actually care about, you're implicitly saying yes to the other 90. That's how governance enables speed rather than blocking it.
She also made a point that doesn't get made enough: governance committees need business stakeholders, not just lawyers. Legal can flag risks. Only the business knows whether a tool is actually a priority.
Patrick Murta from Centene echoed this with a concrete model. All prompts at Centene flow through a centralized AI gateway with full auditability. When regulators ask questions, the team can show exactly what the policy was, when it executed, and what it prevented. That's not just compliance. That's operational confidence.
Scott Litchkin from PwC named something that resonates with anyone who has tried to roll out a new capability inside a large organization. The C-suite is bought in. Newer employees are native to the tools. The resistance lives in the middle — managers who have built expertise around existing processes and have the most to lose from disruption.
The fix isn't more communication from the top. It's showing those managers something specific that would actually change their day. The light bulb moment is personal and concrete, not abstract and strategic.
Ryan Hiser from Lumen offered a practical example of how this plays out in rollout design. They didn't push Copilot to everyone at once. They started with the highest-use teams. Teams without access developed FOMO. That created organic demand. Eventually the common language across the organization mattered enough to justify a company-wide rollout — and by then, people wanted it.
Section's research supports the specific levers that move the needle. Clear tool access processes correlate with a 3x increase in proficiency. Training that covers agents and automations drives a 2.5x increase. And managers actively demonstrating their own AI use correlates with a 2.1x increase. That last one is particularly important: 79% of managers have not shown their team their own AI use in the past month. Change lives or dies at the manager level.
Amal Fadke from Tech Mahindra made a distinction that cuts through a lot of confusion in enterprise data conversations. Data readiness doesn't mean your data is centralized or perfectly structured. It means clear access controls, agreed-upon semantic layers, and single sources of truth for specific domains.
Patrick Murta framed it similarly at Centene. Core payer data and run-the-business data have different governance requirements. The point isn't to standardize everything. It's to know what you have, who can access it, and what it means.
This connects directly to a barrier that slows down AI programs in almost every large organization: teams can't find data assets that already exist, so they build their own. The result is redundant work, inconsistent outputs, and a fragmented foundation that makes real-time action harder with every new project. Discoverability isn't a nice-to-have. It's what makes shared data assets actually get used.
Scott Litchkin was direct about this. A lot of what's being called "agentic" is just RPA with better marketing. True value comes from combining LLM reasoning with deterministic tools and real data access. That combination is what makes an agent meaningfully different from a smarter chatbot.
Centene has two agents already in production worth noting. An estimate builder that draws on historical project data to generate software development estimates in near-real-time, replacing a process that used to take weeks. And a competitive intelligence agent that aggregates competitor medical policies and market signals into a distilled briefing, eliminating manual research across a team of analysts. The prototype was built in under a week.
The lesson isn't that agents are easy. It's that the right use cases, built with the right data and the right governance, can move fast and deliver real value. The failure mode is treating every AI project like a moonshot when some of the best early wins are narrow, well-defined, and completed in weeks.
Greg Shove opened the summit with a framing that stayed relevant through every session: AI inference is currently subsidized. Providers are spending far more than they charge. The advice for enterprise leaders is simple. Take full advantage of this window. Experiment aggressively now, because when inference costs normalize, you'll know exactly where it's worth spending.
The organizations that will be in the best position in two years are the ones building that institutional knowledge right now. Not by running the biggest pilots, but by moving consistently, governing well, and actually getting their workforce using these tools.
That requires infrastructure. It requires governance. It requires someone accountable for the program. And it requires treating adoption as seriously as the technology itself.
Those aren't new ideas. But the summit was a useful reminder that most organizations are still not doing them.
Productive Edge helps enterprise teams build the strategy, governance, and technical infrastructure to move from scattered AI experiments to programs that compound over time. If these themes are ones your team is working through, we'd welcome a conversation.