Over the last few years, foundation models gave us a new way to interact with software: you typed a prompt, the model replied, and then you still had to do the real work—choosing what to keep, where to paste it, how to act on it. We had intelligence, but no way to use it effectively.
That's changing. Intelligence itself is becoming part of the user experience.
We're not waiting for smarter models. We already have models that are good enough for a vast range of real problems. The bottleneck has shifted: the constraint is no longer what the model knows, but what the system lets you do with that knowledge. As I wrote in Why I'm Bullish on 2026, we're entering a phase where AI is becoming foundational infrastructure, not just a feature.
AI is moving up the ladder, from reactive generators of text to systems that can hold context, plan, decide, and execute. This mirrors how humans solve problems. The lowest form of help points out that something is wrong. The highest form notices the problem, understands it, and fixes it while keeping everyone in the loop.
In human psychology, agency is the highest level of competence—not just the ability to act, but the ability to initiate, choose, and shape outcomes rather than simply react to them. That's exactly where AI interfaces are heading. The era of AI as a collaborator with agency is emerging, reshaping what interfaces look like.
From prompts to intent
The first wave of AI products was built around prompts. You asked, the model answered. Then came tools and actions—the system could call APIs, search, or write code. That helped, but the initiative still lived with the human.
Prompting was a temporary interface. It was the easiest way to expose a powerful model, but it puts all the cognitive load on the user. You had to know what to ask, how to phrase it, and how to break your problem into steps.
The next generation is built around intent instead of prompts. You don't explain every step. You express a goal. The system figures out the plan, tracks state, and executes.
When systems can maintain state, observe context, and run multi-step workflows, they stop waiting for instructions. They begin to anticipate. This is how software starts to feel like an assistant instead of a form.
Consider a marketing agent that monitors your campaigns, identifies underperforming ads, pauses them, and reallocates budget—all without you asking. Or a coding agent that watches your codebase, notices when tests start failing, investigates the cause, and proposes fixes before you even see the error. These aren't chatbots responding to prompts. They're operators that understand intent and act on it.
Voice is accelerating this shift. Your car, your phone, your home become surfaces for intelligence. You don't type a command. You speak a need. Sometimes you don't even need to speak—the system infers intent from patterns and environment.
Power users will go further. They'll train their agents with preferences, goals, principles, and constraints, so the system can make decisions on their behalf. Over time, these agents become personalized operators that know how you trade off speed versus quality, risk versus reward, privacy versus convenience. You're not just delegating tasks—you're delegating judgment. It's about compressing the distance between a problem and a result.
This is why agents matter. Not because they're smarter, but because they're allowed to remember, to plan, and to act. They turn intelligence into a usable surface.
From answers to outcomes
Chatbots give you answers. Agents give you outcomes. That sounds like a small distinction, but it's everything. An answer still requires a human to turn it into action. An outcome is already inside a workflow.
When an AI system can read your data, apply rules, call APIs, make decisions, and update state, it stops being a novelty and starts being infrastructure.
Productivity comes from tighter loops between thinking and doing, not from better text.
Interfaces become the bottleneck
Once models are good enough, the limiting factor is no longer intelligence. It's how that intelligence is exposed.
We're already seeing this in practice. The best AI tools aren't the ones with the biggest models. They're the ones with the best flows: where context persists, tasks are visible, progress is tracked, and humans and machines share the same workspace.
The user experience becomes the product. This is the shift that matters most in 2026.
Designing high-agency AI
Good AI UX is about trust, visibility, and control. You need to know what the agent is doing, why it's doing it, and when you can override it. Without that, proactivity feels like loss of control instead of leverage.
The biggest mistake people make about agents is thinking of them as autonomous beings. In reality, they're closer to very powerful macros. They chain steps. They maintain memory. They call tools. They react to changes.
This isn't mystical. It's just software architecture catching up to intelligence.
Once you think of agents this way, a lot becomes clear. You don't need a general super-intelligence. You need well-designed systems that can move through work the way a human assistant would.
Why 2026 feels different
This feels like a real inflection point because three things are finally aligning: models are good enough, tooling is becoming composable, and interfaces are being rebuilt around workflows instead of chat.
That combination turns AI from a feature into a foundation.
We're not entering the era of smarter machines. We're entering the era of usable intelligence.
And that's a much bigger deal. For founders and designers, this means the competitive advantage isn't in having the best model—it's in building the best workflows. For users, it means AI stops feeling like a tool you use and starts feeling like a collaborator you work with. The infrastructure is ready. The question is what we build with it.