Five observations on the enterprise operations of AI
Most of the conversation around AI is still about what the technology can do. I think that’s the wrong thing to be watching. The capabilities are arriving faster than almost anyone planned for, and they will keep arriving, but capability was never going to be the constraint.
What separates the organizations that pull ahead from the ones that stall is something far less glamorous: whether they’ve built a foundation strong enough to put AI to work across the business, safely and at scale.
I spend most of my time now on that second question. And the more I look, the more convinced I am that the foundation is where success with AI gets decided. Here are five things I’ve been thinking about.
1. The right foundation doesn’t add complexity, it unwinds it.
After Build 2026, what stood out to me wasn’t any single capability, it was the discipline. Microsoft is building a complete stack for the AI era layer by layer: infrastructure, identity, context, security, governance. Done well, a foundation doesn’t pile on complexity; it absorbs it. It gives you a stable base to build on, even while the layers above keep changing.
Most organizations are feeling the opposite right now. Complexity is multiplying with every new model, agent, and integration. That sprawl is a signal that the foundation is missing or underbuilt.
The first phase of AI was experimentation; this phase is operations and you can’t operate on the infrastructure equivalent of sand. Get the foundation right and the complexity above it becomes manageable. Skip it, and every new capability just adds more weight to an already shaky foundation.
The action: Follow the complexity. Wherever AI initiatives are slowing down, look underneath. The bottleneck is often not the model or agent: it’s some blend of identity, context, security, or governance. That can point you towards where to strengthen foundations first.
2. Shadow AI is becoming Shadow Operations.
For the last two years we’ve worried about employees using AI tools outside approved channels. That’s still a challenge, but I suspect the bigger issue is what’s coming next.
AI is no longer a copilot sitting beside your people; it’s an agent woven into the fabric of your digital core. Every tenant, every identity, every data boundary in your Microsoft estate now has to assume that agents are acting, not just assisting. How do you secure and manage that across every control plane? There’s no single tool to rule them all.
The action: Explore Agent365, not in isolation, but as just one part of your agent control plane. Organizations will need a portfolio of identity, governance, security, and observability capabilities working together to fully understand where agents exist, what they can access, and how they interact with the environment.
Also Read: AiThority Interview with Gou Rao, co-founder and CEO at NeuBird AI
3. The next competitive advantage may be organizational context.
General-purpose models know a lot about the world. What they don’t know is your business. The organizations creating the most value from AI won’t necessarily be the ones with the best models. They’ll be the ones that are best at grounding AI in the processes, decisions, and thousands of small realities that make your organization unique.
The foundation for this already exists. Tying AI into the rest of your business is the artifacts, domain knowledge, and systems of record where your real institutional context lives. It takes an additional layer of knowledge tailored to your organization as your own IQ layer.
This is the behind-the-scenes, connective-tissue work that we spend much of our own time on. We’ve built a dedicated Microsoft Purview and Fabric data governance solution around exactly this: a governance council, clear ownership and stewardship, a business glossary and catalog, lineage across Azure and Fabric, and data-quality scorecards, with AI-ready controls layered on top.
The action: Start with Microsoft’s IQ layers and follow the trail. As you connect it, you’ll quickly discover where business knowledge is fragmented, undocumented, or poorly governed. Those gaps can become the roadmap for your own organizational intelligence layer.
4. Distribution is fragmenting, and sovereignty is moving to the center.
For most of the digital era, distribution was almost free. You built a product once (like Instagram or Spotify), launched it everywhere, and then made local adjustments for language and culture.
AI is breaking that model. Regulation is starting to shape not just how products launch but whether they launch in a given market at all. The clearest example landed in early June: at WWDC 2026, Apple announced that its rebuilt Siri with Apple Intelligence and it won’t ship on iPhone or iPad in the EU later this year, citing the EU’s Digital Markets Act, with no timeline for when EU users will get it. Same company, same product, now shipping differently depending on where you are.
That’s the visible edge of a deeper shift. For years, governance and sovereignty were things we layered on after the technology decisions were already made. AI is reversing that sequence. Questions about data access, residency, risk tolerance, and human oversight are starting to determine what gets built in the first place and where it’s allowed to run, especially as you add the complexity of agents and pay-as-you-go consumption.
What’s easy to miss is that sovereignty isn’t only a constraint. The organizations that build these capabilities deliberately often find they become a differentiator, even a growth lever, giving them the trust, control, and flexibility to deploy AI more broadly and confidently than competitors who treated it as an afterthought.
The action: Treat sovereignty not as a risk play, but one focused on growth. When you are creating a new product or service and think about it early in design of your products and services.
5. FinOps must evolve down to the individual.
The cloud era taught us that flexibility without governance gets expensive fast. AI is the same lesson, intensified. Every organization will need to evolve its FinOps to manage AI consumption alongside cloud consumption. The tooling is early, just as it was in the first days of cloud, and it will mature. But there’s a real difference this time: cloud cost management lived mostly at the org and team level, while AI consumption comes down to the individual, and who’s using which models, how often, and how well. Managing that takes a different set of tools and a different mindset: not just controlling spend but helping people use AI effectively rather than simply using more of it.
The action: Don’t wait for the tooling to mature. Start measuring AI consumption today, even if it requires a patchwork of approaches and dashboards. Keep an eye on the startup ecosystem; many of the tools in this space are still being built.
The next chapter is about what organizations can do with AI—and that gets decided in the foundations, not the pilots. If you can’t answer who governs your agents and what they cost, you have AI experiments. Building the foundation may be the least glamorous part of this story, but it’s the part that matters most.
Also Read: AI and The Future of Work: Artificial Intelligence Is Expanding Organizational Intelligence Beyond Human Limits
[To share your insights with us, please write to psen@itechseries.com]
Comments are closed.