Five Things I Learned at Ai4 2026 (Part 1)

If you attend enough technology conferences, you start to recognize the pattern. Every year brings a new buzzword. Every keynote promises a coming revolution. Every vendor claims to have solved the industry’s most pressing problem.

Ai4 2026 certainly had its share of typical conference hype. With more than 12,000 attendees, it has become one of the largest independent AI conferences in the world, and there was no shortage of product announcements, platform demos, ambitious predictions, and vendors eager to explain why their platform represented the future of enterprise AI.

But after two days of moving between sessions, I found myself rewriting my notes.

What initially looked like dozens of unrelated presentations gradually coalesced into a handful of recurring themes. Whether the speaker came from healthcare, financial services, marketing, Google Cloud, the NFL, or an AI startup, the conversations kept circling back to the same ideas. It was as if the industry had quietly moved on from debating what AI can do and started wrestling with what organizations must do to make AI actually work. That may have been the biggest takeaway from the conference.

AI itself is no longer the story. The real story is everything organizations have to now build around it.

AI Is Becoming an Operating Model, Not a Tool

The first pattern emerged almost immediately. Across multiple sessions, speakers described AI less as a technology initiative and more as an organizational transformation. They weren’t talking about choosing the right model or buying another platform. They were talking about redesigning how work gets done.

Dataiku made this point especially well. Rather than framing unauthorized AI usage as a compliance problem, the presenters argued that it should be viewed as a signal. Employees are trying to solve legitimate business problems faster than the organization is equipped to support them. The demand is real. The infrastructure is missing. Instead of asking how to stop people from using AI, organizations should be asking how to build the foundation that allows them to use it responsibly.

That same theme surfaced again during a fireside chat with the NFL. Their speakers, Celeny Da Silva and Chi Ogbuehi, described AI not as a technology journey but as an operational transformation. Before scaling AI, they had to rethink people, processes, metadata, governance, and operating models. They emphasized identifying champions, standardizing how information is described, and establishing clear workflows before expecting AI to deliver meaningful business value.

Jed Dougherty and Mark Abramovitz from Dataiku summarized the challenge with a line that stayed with me for the rest of the conference:

“This is an organization design problem wearing a technology costume.”

That observation echoed throughout the event. Whether the discussion centered on marketing, healthcare, software development, or enterprise IT, the conversation repeatedly returned to organizational foundations. Companies aren’t struggling because the models aren’t capable enough. They’re struggling because operating models built for software don’t automatically work for intelligent agents.

There was another subtle shift that felt significant. A year ago, much of the AI conversation revolved around experimentation. This year, speakers were talking about ownership. Who is responsible for an agent? Who measures its success? Who approves its actions? Dataiku even suggested organizations should begin thinking about agents the way they think about employees, each with an owner, a budget, a purpose, and measurable outcomes.

The implication is hard to miss. AI isn’t becoming another application in the enterprise stack. It’s becoming part of the operating model.

The Knowledge Layer Is Becoming the Competitive Advantage

If there was one phrase that echoed throughout the conference, it was some variation of “the knowledge layer.” Different speakers used different terminology, but they were all describing the same phenomenon: AI is only as valuable as the information it can retrieve, understand, and trust.

Wiley’s presentation on trust in AI argued that organizations are drowning in information while becoming increasingly starved for insight. Their research suggested the real challenge isn’t model intelligence, it’s knowledge quality. If inaccurate or outdated information enters the retrieval layer, later reasoning steps can’t magically repair it. In agentic workflows, mistakes don’t simply persist, they compound as agents build one decision on top of another.

PromptQL approached the same problem from a completely different angle. Their vision centered on capturing what already exists inside people’s heads and transforming it into a shared “company brain.” Instead of every analyst rediscovering the same knowledge, teams teach the system once and allow that expertise to become reusable across the organization. The emphasis wasn’t on replacing human knowledge. It was on preserving and scaling it.

Coral extended that argument to enterprise agents. Their case for building agents in-house wasn’t simply about lowering costs or avoiding vendor lock-in. It was about ownership. When your organization’s knowledge, retrieval strategies, and institutional context live inside someone else’s platform, you’re helping build their strategic asset instead of your own.

Even Pfizer’s presentation on its CoCo assistant reflected the same philosophy. Rather than positioning AI as a better dashboard, the team focused on building a knowledge-graph-powered system that could connect disparate commercial signals, explain findings, assemble evidence, and recommend next actions for human review. The intelligence didn’t come from the language model alone. It came from the quality of the knowledge surrounding it.

Different industries. Different products. Different speakers. Yet the same conclusion kept surfacing: the competitive advantage is no longer the model itself. It’s the quality, structure, and ownership of the knowledge surrounding that model.

Marketing Is Returning to Marketing

One of the biggest surprises came from the marketing track. Going in, I expected to hear a steady stream of presentations about AI-generated campaigns, synthetic content, and autonomous marketing. Those topics certainly came up. But they weren’t the focus.

Instead, speaker after speaker argued that AI is removing production work while making human judgment even more valuable.

One panelist, Monica Ho, CMO of SOCi, described using AI to compress go-to-market timelines from months to weeks, but only after investing heavily in the foundational work needed to train the system and establish oversight. Another explained how AI had transformed content production, yet emphasized that the real bottleneck had shifted to strategy, messaging, and decision-making. Several organizations discussed using AI to localize content, personalize campaigns, or monitor hundreds of websites, but they consistently framed these capabilities as freeing marketers to spend more time understanding customers and less time executing repetitive tasks.

Perhaps the most memorable comment came near the end of one panel. Christy Borrowman, VP of Marketing & Transformation at Colgate-Palmolive, quoted Rebecca Hinds, Ph.D., from Glean:

“Don’t automate the soul out of the work.”

The room immediately seemed to understand what the speaker meant.

Marketing has always balanced science and art. AI appears to be taking over much of the science, the production, the optimization, the repetitive execution. That leaves marketers with a greater responsibility to provide the things machines still struggle to produce: judgment, empathy, narrative, taste, and an authentic understanding of customers.

Another observation surfaced repeatedly. Several marketing leaders suggested that the era of “growth at all costs” is fading. AI can help teams move faster, but speed alone is quickly becoming table stakes. The differentiator shifts back to brand, trust, creativity, and customer relationships, the very qualities that defined great marketing long before generative AI existed.

In an odd way, AI may be pushing marketing back toward its fundamentals.

Enterprise AI Is Really an Architecture Problem

Perhaps the biggest surprise of the conference was what people weren’t talking about.

There was remarkably little discussion about which foundation model would ultimately win. Few sessions focused on prompt engineering. Benchmark scores barely entered the conversation. Instead, everyone seemed preoccupied with the architecture surrounding the models.

Context.

Memory.

Governance.

Semantic layers.

Knowledge graphs.

Policy.

Observability.

Shared metadata.

Retrieval.

These ideas appeared over and over again, regardless of industry.

One presentation described this surrounding infrastructure as the “chassis” of enterprise AI. Models are becoming increasingly interchangeable. What matters is the context they receive, the tools they can access, the policies that constrain them, and the evidence supporting their conclusions. Enterprise-grade AI, the speaker argued, isn’t a bigger model. It’s a better chassis.

Another presenter observed that organizations don’t actually have a tooling problem. They have a connective tissue problem. Marketing teams may have dozens of excellent platforms, each solving its own piece of the puzzle, but nobody has stitched them together into a coherent system. Humans quietly became the integration layer, spending enormous amounts of time moving information from one tool to another instead of creating value.

That architectural perspective appeared everywhere. Conversations about semantic layers surfaced during discussions of conversational analytics. Metadata appeared in talks about operational transformation. Knowledge graphs emerged in pharmaceutical AI. Shared context became central to enterprise agents.

The common thread wasn’t AI, but systems design.

Everyone Is Still Figuring It Out

For all the confidence on display from vendors and keynote speakers, one observation kept resurfacing in my notebook. Most people are still figuring this out. That wasn’t meant as criticism. If anything, it was reassuring.

The organizations presenting on stage weren’t claiming to have reached the finish line. They were openly discussing pilot projects, organizational resistance, governance challenges, data quality issues, and lessons learned from things that hadn’t worked the first time. There was far more honesty than hype.

Several sessions even challenged assumptions that have become common in AI conversations. Unauthorized AI use wasn’t always treated as a security problem. It was often described as evidence of unmet demand. Slow adoption wasn’t blamed on reluctant employees so much as weak foundations. Technical limitations were frequently overshadowed by organizational ones.

By the end of the conference, I found myself with a very different impression than I expected. I arrived expecting to hear about increasingly capable AI models. I left thinking much more about organizations.

The recurring themes weren’t faster models or bigger context windows. They were shared knowledge, governance, operating models, architecture, trust, and human judgment. Those topics surfaced so consistently across industries that they began to feel less like individual observations and more like a snapshot of where enterprise AI is heading.

If there was one lesson that tied the conference together, it might be this:

AI isn’t replacing the fundamentals of good organizations. It’s making those fundamentals impossible to ignore.

Coming Next

This article focused on the broad themes that emerged across Ai4 2026. In Part 2, I’ll shift from observation to action by looking at the practical advice speakers repeatedly offered. We’ll explore why documentation is becoming strategic infrastructure, how organizations are building “company brains,” what it really means to own enterprise AI agents, and why the companies making the most progress are spending at least as much time on foundations as they are on models.