What is the single most important AI capability that will become mainstream in the next 12 months?
That most important capability splits by audience; and consumers move first because they tolerate more risk. For individuals, computer use is the capability: AI that acts through the browser and desktop rather than just talking in a chat window. It's already happening, and over the next year delegating real tasks (booking, shopping, routine admin) end to end becomes normal. The enterprise rides the same wave but will be a step behind, because the bar for trust is higher. There, the capability that actually unlocks agentic success is long-horizon memory. Enterprise work spans days, systems, and people, not a single session, and agents that forget stay stuck as demos. Persistent context is what turns an agent from a clever prototype into something upon which a business can depend. Same trajectory, different clock speed.
How will AI change the way organizations make decisions by 2027?
The shift is from decision support to decision execution. Today AI informs a human who decides. By 2027, agents will increasingly draft the decision and the evidence behind it, with humans moving into an approval and oversight role. That sounds like a small change, but it inverts the workflow. What makes or breaks that shift is trust, and trust is earned through visibility. Organizations won't accept agentic decisions they can't inspect. The ability to see the agent's trail (what it looked at, why it concluded what it did, where a human intervened) becomes a fundamental requirement, not a nice-to-have. Reliable, auditable reasoning is the price of admission. So the real change isn't just faster decisions; it's that governance and traceability stop being compliance overhead and become the thing that determines whether AI gets adopted at all.
What AI risk or challenge is most underestimated right now?
Everyone knows the challenges by name: the evaluation gap, agent sprawl, over-trust in high-stakes decisions. What's underestimated is the one underneath all of them: the foundation. The data, integration, and governance work that nobody puts on a keynote slide. It's underestimated because it's hard to scope and easy to defer, and because its cost shows up somewhere else. Underinvest in the foundation and evaluations get harder, because you can't measure correctness against data you don't trust. Agent sprawl gets worse, because there's no shared control plane for agents to build on. Over-trust grows, because no one can trace what the agent actually relied on. So the foundation isn't one risk among four. It's the root cause that quietly determines how bad the other three get. It's not a model problem. It's a foundation problem.
What advice would you give to a company starting its AI journey today?
Start with your information, not the model. The models are largely commoditized and getting better on their own. What you actually own, and what no competitor can replicate, is your information estate: your documents, your data, your institutional knowledge. That's the moat, and it's also the fuel. AI is only ever as good as the information you point it at. So the first move isn't picking a model or a use case. It's getting your information in order: secured, governed, and accessible to the systems that will use it. Most teams skip this because it's less exciting than a demo, then wonder why the demo never scales. Get the information foundation right and the use cases get easier, the outputs get more trustworthy, and every later choice gets cheaper. Secure information management isn't the boring prerequisite to AI. It's the thing that makes AI work.
What AI trend are you most excited about personally?
What excites me is the harness maturing, not the model. The model stopped being the hard part. What's getting good fast is everything around it: connectors, memory, data access, the plumbing that turns a model into an assistant that actually does things. I'm most excited watching this as a consumer. Individual users are quietly offloading real productivity work to AI: recommendations, booking, even filing taxes, handled end to end. And they're doing more of it every month. The pattern is consistent: harness, data, memory, connectors. That's the recipe, and consumers are proving it first. The exciting part is that this consumer blueprint is exactly what the enterprise needs next. Same recipe, harder kitchen.
