Developer conference season has a familiar rhythm: big announcements, bold demos, and a steady drumbeat of “this changes everything.” This year’s refrain is even louder—and more specific. Instead of simply promising faster chips or smarter software, the industry is pitching a new way to use your computer entirely, with AI at the center of the experience. Nvidia is leading with the most literal interpretation of that idea: a laptop designed around AI from the ground up. Microsoft and Google, meanwhile, are pushing the software side—agents, new workflows, and interfaces that assume you’ll delegate tasks to models rather than manually orchestrate every step.
The question hanging over all of it is the same one people keep asking about AI products in general: does anyone actually want this? Not in the abstract sense of “AI is cool,” but in the practical sense of “will this make my day-to-day computing meaningfully better, reliably, and soon enough that it feels worth switching?” The answer may depend less on how impressive the demos are and more on whether these systems can earn trust, reduce friction, and fit into real habits—especially on laptops, where users expect responsiveness, privacy, and control.
What Nvidia is really selling isn’t just performance—it’s a new interaction model
Nvidia’s Jensen Huang made the company’s position unusually clear this week: the next leap isn’t merely incremental upgrades to existing laptops, but “a completely new way of using our laptops.” That phrasing matters. It implies that the hardware platform is being redesigned to support a different kind of workload and, crucially, a different kind of user expectation.
In the past, laptop upgrades were mostly about speed and efficiency. You bought a better CPU or GPU because it made your current tasks faster: editing video, rendering 3D, running local models, gaming at higher frame rates. Even when AI entered the picture, it often arrived as an add-on feature—something you could try if you wanted, not something that fundamentally changed how you interact with the machine.
A platform built for AI changes the premise. It suggests that the laptop will be optimized for continuous inference, low-latency responses, and perhaps a more integrated pipeline between the device and the model. That could mean dedicated acceleration for AI workloads, memory and power configurations tuned for model execution, and system-level features that treat AI as a first-class component rather than a background process.
But the deeper shift is the implied workflow: instead of you typing prompts and waiting for results, the laptop becomes a collaborator that can interpret intent, plan steps, and execute actions across apps. In other words, the laptop isn’t just a screen for AI output—it’s the environment where AI operates.
That’s exciting, but it also raises a hard requirement: the system must behave predictably enough that users feel comfortable letting it act. If the AI is constantly wrong, slow, or confusing, the “new way” becomes a new source of frustration. Laptops are personal tools. People tolerate experimentation on phones and desktops in controlled ways; they’re less patient when the device is the hub of their work, communication, and daily logistics.
The “new kind of laptop” pitch is also a bet on on-device and accelerated AI
Nvidia’s framing aligns with a broader industry trend: moving more AI capability closer to the user. There are two reasons companies keep returning to this idea.
First, latency. If every meaningful action requires sending data to a distant server, the experience will always feel like a remote service. Even if the model is brilliant, the delay can break the illusion of fluid collaboration. On-device or accelerated inference can make interactions feel immediate—closer to how you expect software to respond.
Second, privacy and control. Laptops contain sensitive information: documents, messages, browsing history, work credentials. Users don’t want to assume that everything they do is being shipped off to the cloud. While cloud AI can be secure, the perception problem is real. A laptop that can handle more locally can reduce that anxiety and make AI feel less like surveillance-by-default.
However, there’s a tradeoff. Running more locally can increase cost, complexity, and power demands. It also forces manufacturers to decide what “AI-first” means in practice: which tasks run on-device, which are offloaded, and how the system manages the boundary between them. If that boundary is opaque, users may experience inconsistent behavior—fast in one scenario, sluggish in another, or mysteriously limited when the model can’t access enough compute.
So the success of an AI-focused laptop depends on orchestration. The hardware is only half the story; the software stack has to make the experience coherent.
Microsoft and Google are pushing agents and workflows that assume delegation
While Nvidia is talking about a new laptop platform, Microsoft and Google are emphasizing the software layer—especially agents and workflows that treat AI as an active participant.
At Microsoft Build and Google I/O, the recurring theme is that AI won’t just answer questions. It will help you plan, coordinate, and carry out multi-step tasks. That’s the agent vision: systems that can interpret goals, break them into steps, use tools, and iterate until the task is done.
This is where the “does anyone want this?” question becomes more nuanced. Many people say they want AI assistance, but they often mean one of two things: either quick answers (like a smarter search) or content generation (summaries, drafts, translations). Agents are different. They require the user to trust the system with actions that can have consequences—sending emails, editing documents, booking travel, changing settings, or pulling data from multiple sources.
Agents also introduce a new kind of cognitive load. If the AI is doing the work, users still need to understand what’s happening enough to correct course. That means the interface has to communicate intent, progress, and uncertainty clearly. Otherwise, delegation turns into a black box: you asked for help, but you can’t tell whether the system is working toward the right outcome.
The best agent experiences will feel less like “AI magic” and more like “a capable assistant with guardrails.” The worst ones will feel like a chaotic intern who keeps trying, but never quite gets it right.
The Vergecast angle: the industry is converging on AI as the operating layer
On the latest episode of The Vergecast, Nilay and David run through a range of products coming out of Microsoft Build and Google I/O, including directions for AI agents and the broader push toward AI-shaped computing. The conversation reflects what’s happening across the ecosystem: companies aren’t just adding AI features; they’re trying to reposition AI as the layer that sits between you and your tools.
That’s a subtle but important shift. Traditional software is built around explicit user actions: click here, type there, run this command. AI-shaped computing aims to replace some of those explicit steps with intent-based interaction. You describe what you want, and the system figures out how to get there.
This is where laptops become the battleground. Phones are constrained by screen size and input methods; desktops are powerful but often fragmented across apps and workflows. Laptops are the middle ground where people do serious work and expect both flexibility and reliability. If AI can truly improve laptop workflows—writing, organizing, researching, coding, managing schedules—it could become indispensable. If it can’t, it risks becoming another layer of novelty that people disable after the initial curiosity fades.
A unique take: the real product isn’t the model—it’s the friction profile
It’s tempting to evaluate AI laptop pitches by asking whether the model is smarter than last year’s. But the more meaningful metric might be something else: the friction profile.
Every AI workflow has friction points: prompting, waiting, verifying, correcting, and integrating results back into your existing tools. Even if the AI output is good, the workflow can still be painful if it requires too many manual steps to make the result usable. For example, a summary that can’t be easily pasted into the right document, or a plan that doesn’t connect to your calendar, or a code suggestion that doesn’t match your project context—these failures don’t just reduce quality; they increase the time and attention required from the user.
An AI-first laptop succeeds if it reduces friction across the entire loop:
1) Understand intent quickly.
2) Execute steps with minimal back-and-forth.
3) Present results in a form that fits your workflow.
4) Make it easy to verify and correct without starting over.
This is why the “completely new way of using laptops” claim is both promising and risky. Promising, because if the system is designed end-to-end, it can streamline the loop. Risky, because if the loop is still clunky, the new hardware becomes a costly wrapper around the same old experience.
The friction profile also determines whether people “want it.” Most users don’t want to learn a new interaction paradigm unless it saves time or reduces stress. They’ll tolerate a learning curve if the payoff is immediate and consistent. They won’t if the payoff is occasional or unpredictable.
Where the market could land: AI laptops as specialized assistants, not universal replacements
One possibility is that AI laptops won’t replace traditional computing so much as they will create a new category of “assistant mode.” Think of it like how browsers evolved from simple pages into platforms with extensions, permissions, and integrated services. Over time, the browser became the environment where many tasks happen. Similarly, the AI laptop could become the environment where certain classes of tasks are handled automatically.
In that scenario, the laptop wouldn’t need to be perfect at everything. It would need to be excellent at a few high-value workflows—ones that are frequent, time-consuming, and repetitive. Examples include:
– Turning messy notes into structured drafts.
– Summarizing long threads and extracting action items.
– Helping plan trips or meetings with constraints and preferences.
– Assisting with coding by understanding project context and proposing changes.
– Managing personal knowledge: finding relevant documents, comparing versions, and updating records.
If AI can own those workflows reliably, users will adopt it because it feels like a