Code, Context, and Capacity: Aligning Your AI Product and Infrastructure Strategy | Domo BUILD 2026

So without further ado, I want to introduce our first presenter here today and his name is Jim Fairweather. Jim Fairweather is a long-time leader and he is the Head of AI Go-To-Market at Google. He is going to join us to show us how we can align our infrastructure strategy with our product strategy to build enterprise-grade agentic solutions. For him, the future of AI isn't just about better prompts—and we're all learning how to write better prompts—it is where and at what scale code runs. So with that, please welcome Jim Fairweather.

Thanks so much, Matt, and I will take over the screen share here. Appreciate that intro and hello to everybody out there.

If you look at telemetry data of a Formula One racing team, you won't find discussions about the color of the paint or the sponsor logo on the driver's suit. You'll find discussions about capacity, downforce coefficients, brake thermodynamics, and fuel flow rates. In the world of high-performance racing, your output is strictly bound by your underlying capacity. If you have the most beautiful carbon fiber chassis in the world, the most elegant aerodynamic code in your bodywork, but the engine lacks the physical capacity to deliver the raw horsepower, you'll be left sitting useless on the starting grid.

Today, the technology world is facing its own engineering crisis on the starting grid. For the last 18 months, our conversations have been obsessed with code: which model has the most parameters, which one has the longest context window. While the code is essential, you can't win a Grand Prix in a minivan. The engine alone does not win the race. The AI model is just a component. It's completely blind without context and completely useless without the physical capacity.

I work with software and technology companies across the country, and one common mistake I see is what we call the model-only illusion. We're seeing this commoditization of brilliance where the models are getting smarter, faster, and cheaper every day. But here's the hard truth: code is a level playing field, not a competitive advantage if everyone has access to the same high-powered engine. The engine isn't why you win; you win because of how you use it. Frankly, we've seen some misuses of these powerful coding models, and they show up rather clearly on the balance sheet.

We must move past this idea of one model to rule them all. The winning strategy for the next decade is a portfolio play. You don't run your hybrid engine at 100% full-throttle performance when you're cruising down the pit lane. It's a waste of capacity and fuel. You don't need the most powerful models to summarize emails. In the agentic era, you must route your tasks dynamically. You route your simple, high-speed agentic actions to lightning-fast, ultra-low-cost models like a flash model. You save your state-of-the-art models like a pro model for the heavy, multi-step logical reasoning. And you leverage native models like a Gemini Omni when you need a seamless blend of text, audio, image, and video, editing the content step-by-step using natural language. For highly proprietary, sovereign data, you deploy open-weight models like Gemma or others in your virtual private cloud. That's model redundancy. It's about building a modular abstraction layer so that if a model has a bad Friday, your system doesn't even blink.

To win the race, the driver needs to have real-time telemetry. You need context, and part of that context is how ready the organization is to take on AI agentic developments evolving at an exponential pace, to avoid scaling more chaos than innovation. There are three best practices we've seen clients use to bring rigor to their agentic strategy.

The first is to prioritize with precision. The most common pitfall I see in organizations is that they treat AI as a solution in search of a problem. We want to identify where the most friction exists and where we can make the most impact on a critical business outcome.

The next is to build a unified data foundation. AI is only as powerful as the data that feeds it. Data hygiene isn't an IT task; it's a foundational business imperative that creates a sustainable competitive advantage. Our friends at Domo know a thing or two about this.

Lastly, you have to activate the organization. Technology alone does not create value; people do. It demands a cultural shift, upskilling, and leadership. We must securely ground our models natively into our enterprise data, our databases, our document stores, and our real-time searches, because we're transitioning from instruction-based computing, where we click buttons and write lines of code, to intention-based computing, where we state a goal and an agent executes it.

There is one thing I'd like to call out: building individual productivity is not the same thing as building institutional value. The tooling and the governance approaches are vastly different, and there's a time and a place for each. In fact, institutional AI should take inspiration from individual AI. However, many organizations are dealing with agent sprawl, where the creativity and chaos of individual AI is bleeding into institutional AI, causing confusion for users, increasing organizational risk, and presenting a nightmare to govern.

There needs to be a balance of creativity and speed against coordination and safety. It's a challenge many IT teams are facing, and it should sound familiar: a demand for more tools than they have the capacity to build, inconsistent tracking, different user interfaces and styles making it difficult to learn, use, and navigate, and a ton of duplication. We want to think about it slightly differently. Instead of thinking about it as governance for teams or people who want to build and deploy agents, we like to think about it as lighting a path for agent development.

Building enterprise agents is a lot more complicated than building individual agents. The hardest part about building institutional enterprise AI is the beginning and the end. If you're going to resource this, you need to focus on defining what you are building, who the user is, what the ideal user experience should look like, and when they would use it.

Once you've built something, how do you measure the ROI? It's a common question coming up again and again, especially from the board. This is a framework we go back to repeatedly when answering these questions.

The first step is to start with AI utilization, which has become table stakes at this point. This is where most companies measure adoption, usually from when individuals save time, and it represents individual AI.

The second step is AI depth. This is measuring how an end-to-end workflow you have could be automated. When we automate workflows, we eliminate bottlenecks. This is where true AI starts to reside and show up within an organization, not just as an individual tool.

The third step is about AI breadth, and this is where the game changes. This is where you're building autonomous operations. You can build systems that don't just monitor the business, but act on it. They see, they hear, and they resolve issues before they happen, creating a self-correcting operation.

Most people aren't here yet, but this is also where the physical reality of the universe catches up with us. Running millions of tokens of context and orchestrating thousands of background agents requires an enormous, unyielding amount of physical capacity. AI doesn't live in the cloud; AI lives in silicon. It lives in liquid-cooled copper pipes, high-bandwidth fiber optic lines, and massive physical concrete structures humming with electrical currents. Every single token your agent generates has a literal physical cost in watts, amps, and thermodynamics. If your product strategy doesn't include an explicit and co-designed infrastructure strategy, you're building a beautiful sports car with a garden hose for a fuel line—it will choke.

How do we bring this together? How do we build internally and externally in this agent era? It requires breaking down the strategic silos in our C-suites. Too often, the product team is busy designing beautiful carbon fiber bodywork, the engineering team is in another room trying to make the engine compile, and the finance team is staring at the fuel bill in absolute horror. We must align. The Chief Product Officer, Chief Technology Officer, and Chief Financial Officer must sit at the exact same table, sharing a single unifying metric: cost per outcome, not just cost per token. They must design AI products around the physical realities of compute capacity, building for model redundancy, securing enterprise grounding as the moat, and treating infrastructure as a first-class citizen of product design.

The gold rush era for AI experimentation is winding down, and the era of operational excellence is officially here. The companies that win the next decade won't be the ones that wrote the loudest, most hype-filled press release; they'll be the ones that had the discipline to align their code, their context, and their capacity. They'll be the builders who designed resilient, cost-effective, and secure digital assembly lines that solve real human problems full speed ahead. Thank you.

‍