Two years ago, the AI industry was obsessed with models. Whose was the smartest? Who topped the benchmarks? Who had the biggest context window? We’re even known for covering the occasional release.
Every time a new model launched, it came with a set of metrics proving it had crushed the competition at reasoning or coding or math. There was also the context-window arms race: Anthropic jumped Claude from 9,000 to 100,000 tokens in 2023. And, just a few months later in February 2024, Google upped the ante with a 1 million-token window for Gemini 1.5.
While that race is still being run, I think a more important one has already kicked off.
Now, every major AI company is fighting to become your assistant.
OpenAI has ChatGPT, Work, Workspace Agents and most recently Dots, its always-on agents. Google has embedded Gemini throughout Workspace and, one year ago, launched Gemini Enterprise. Meta has Muse, a personal agent that follows you across apps, devices and your physical environment. Microsoft has transformed Copilot into an agentic layer across the Microsoft ecosystem. Meanwhile, Anthropic expanded Claude well beyond chat with Cowork and Claude Code. And SpaceXAI has Grok Bot, which gives users persistent AI “teammates” with their own computers that can also work across apps and keep going when you walk away.
These products all look different and have unique pros and cons, but the goal is basically the same: They all want to become the place where you start your workday.
For the winners, that’s an enormously valuable position.
Context Is the Moat
There is another reason everyone wants to own the assistant: context compounds.
An assistant that knows your files is useful. An assistant that knows your files, coworkers, customers, preferences, projects, history and how you like work done is dramatically more useful.
Most importantly, it’s dramatically harder to leave.
If my assistant has spent two years learning how I work, knows my company, has access to all my tools, contains my agents and automations, and handles dozens or hundreds of tasks for me every day, switching assistants starts looking painful.
That’s lock-in.
From Chatbot to Front Door
Think back to how most of us used software four years ago and, frankly, how the vast majority of people still use it today.
I needed to send an email, so I would open Gmail. I needed information from a document, so I opened Drive and searched for what I thought was maybe the document’s name. I needed to update a customer record, so I logged into Salesforce and clicked around until I found what I needed. I needed to research something, so I opened a Chrome browser.
The assistant permanently alters all of those behaviors.
Now, instead of navigating to and through various applications, I tell the assistant what I need: “Find the latest numbers, update the forecast and send Sarah a note explaining what changed.”
The assistant is now in charge of deciding which systems to involve.
The result is that I now care a lot less about any individual software platform, with their varying interfaces and workflows. They’re still essential infrastructure, but they’re no longer where my work begins or where I spend any significant time.
And that is very good news for OpenAI, Google, Meta, Microsoft, Anthropic and/or SpaceXAI. It also helps to explain why the competition has become so intense.
So who’s best positioned to win? Here’s my breakdown of the six leading AI assistant platforms, the advantage each one starts with and where each could fall short.
Six Different Ways to Win
All these companies want to be your go-to assistant, but they’re approaching it from very different positions, which is a big part of what makes all of this so interesting.
OpenAI: The Independent AI Work Layer
OpenAI gave us the first mass-market chatbot of its kind, which provides a huge market advantage — ChatGPT is a destination for so many people who explicitly want to use AI.
Now OpenAI is trying to expand the behavior from “ask ChatGPT a question” to “give ChatGPT the job.” ChatGPT Work can operate across apps and files for extended projects, and Workspace Agents can perform recurring workflows. Dots pushes the idea even further: proactive, always-on agents with their own cloud computers that learn your preferences and connect to thousands of applications through plugins. Matt dug into Dots in this video.
Where it wins: OpenAI is the best choice for people and businesses that want AI itself to become the work environment rather than simply adding AI to the existing software stack.
Advantage: OpenAI isn’t beholden to one platform or suite, so it can position itself as the intelligent layer across everything.
Challenge: It has to earn access to the context (and the enormous enterprise contracts) that Microsoft and Google already have.
Google: The Assistant that Already Knows Where Everything Is
Google’s advantage is almost the reverse of OpenAI’s. If you’re like me, Gmail, Calendar, Docs, Drive and the rest of Workspace already contain an extraordinary amount of your worklife. And that’s not even taking my personal life into account — two decades of emails, all my financial data, ideas for businesses, the occasional love letter. It’s all in Drive.
Gemini can increasingly act across all of those apps, and it’s positioning Gemini Enterprise as the “front door” for workplace AI, connecting enterprise data, workflows and agents. Google also has Gemini Spark, a 24/7 personal agent integrated with Workspace.
Where it wins: Organizations already heavily invested in Google Workspace, and there are many. I can see this being especially compelling to companies that want AI but don’t want to build entirely new workflows for their employees’ day-to-day tasks.
Advantage: Context, and no switching pain: Google doesn't have to convince a Workspace customer to move their email, documents and calendar somewhere new.
Challenge: Making Gemini feel like one coherent assistant rather than AI features tacked on to various legacy Google products.
Meta: The Assistant that Wants to Follow You Into the Physical World
Of these companies, Meta has the most audacious strategy.
Muse is a personal AI agent that runs on its own secure virtual computer and can work across connected apps, remember information about you and take actions on your behalf.
The more interesting part of Meta's strategy may be where Muse is going next.
Meta is bringing Muse to its AI glasses, potentially giving the agent access to what you're seeing as you move through your life. It’s also launched Muse for Small Business, with connections to a well-curated list of essential tools including Shopify, QuickBooks, Slack, Stripe, Canva, Notion and Meta's own Facebook and Instagram business products.
And, Meta has announced a broader Enterprise Platform that will bring Muse, Meta Business Agent, Muse Code and other AI products to business customers.
That's a fascinating strategy. Microsoft and Google begin with your workplace. Meta is making the case that it can begin with your physical body.
Where it wins: Consumers, entrepreneurs and small businesses, particularly those whose business models center on social media, messaging and e-commerce.
Advantage: Distribution on a staggering scale. Meta already owns many of the places where billions of people communicate, discover things and interact with businesses. It also has a hardware path through AI glasses that none of the traditional enterprise software companies can easily replicate.
Challenge: Trust; they’re fighting against some reputational headwinds. To become truly useful, an assistant needs access to some of the most personal information you have. Meta has gone to considerable lengths to build privacy and permission controls into Muse, but convincing people to give Meta that level of access may be a harder sell than building the technology itself.
Microsoft: The Enterprise Control Tower
When it comes to selling into enterprises, Microsoft has the obvious advantage, for better and for worse.
Copilot sits alongside the Microsoft stack that millions of companies have depended on for decades, and Microsoft is increasingly treating agents as an enterprise resource that needs to be deployed, measured, governed and paid for. Its newest Copilot capabilities include long-running agentic work, model choice and centralized controls for managing AI usage and spending.
Where it wins: Large enterprises, particularly existing Microsoft shops where security, governance, procurement and centralized administration matter at least as much as the assistant itself.
Advantage: Distribution. Microsoft doesn't need to introduce itself to the CIO.
Challenge: The exact thing that makes Microsoft such a powerhouse also makes it feel unwieldy. Individual users and smaller teams are more likely to gravitate toward products that feel faster, simpler and cheaper.
Anthropic: The Power User's Workbench
Anthropic’s approach is especially strong for technically sophisticated knowledge work.
Claude Code has become a serious tool for developers, while Cowork (and its attractive interface) brings the same agentic model to broader knowledge work: managing files, operating across applications and completing multi-step tasks. Anthropic has also made impressive inroads in large enterprises, with customers like PwC — the second-largest professional services firm in the world — deploying Cowork and Claude Code across teams.
Where it wins: Developers, technical teams and organizations with complex knowledge work, particularly those with the appetite to invest in and build sophisticated AI workflows.
Advantage: Anthropic has built enormous credibility around Claude as a tool for serious work.
Challenge: Anthropic is directly competing against companies like Microsoft and Google that already own enormous consumer and enterprise contracts.
Grok Bot: Just Give It the Job
This is the one I use every day.
What I like about Grok Bot is almost embarrassingly simple: it is easy.
You create a Bot, give it a job and let it work. Each Bot has a persistent cloud computer, memory, files, browser sessions and tools. Bots can work while your laptop is closed, learn a workflow after you show them how to do it and coordinate with other Bots. Matt walked through Grok Bot here.
A surprising amount of my work has moved there.
And the thing that keeps me using it isn't that it never gets stuck. It does. But I’ve always found a straightforward path to fixing the issue permanently. I rarely feel like I'm hitting the same wall over and over.
Not frustrating users matters more than another benchmark win.
Where it wins: Power users, founders, small teams and operators who want to delegate real work without first becoming experts in agent architecture.
Advantage: The mental model is incredibly intuitive. You're giving work to a teammate, not constructing an automation.
Challenge: It is newer than the major enterprise platforms and still has to prove that this simplicity can scale across large organizations, although SpaceXAI has already begun adding enterprise governance and shared Team Bots.
The New Platform War
As AI assistants become mainstream, they create an interesting reversal: AI may not destroy platform lock-in. It may create an even stickier layer above today’s software stack.
That’s why this race matters so much. These companies aren’t just competing to build the smartest assistant. They’re competing to become the layer where your work actually happens.
And that’s not a criticism. All of these companies are businesses, and they need to make money. Becoming indispensable to your customers is a pretty good business model.
The economics could also get very attractive as agents become more capable. I might ask a chatbot five questions a day. A genuinely useful assistant could participate in dozens or hundreds of tasks: sorting emails, tracking news, reformatting content, and updating documentation.
With every new responsibility an assistant takes on, the more valuable it becomes for both the user and, crucially, for the assistant’s developer.
For all of these companies, the assistant is a wedge designed to capture our attention, but the real prize is owning the interface through which people work, and with it, the distribution, customer relationships and recurring revenue that follow.
Most consumers don’t care about whether one company’s model scores two points higher than another’s on a benchmark, but most of us care a lot about products that are easy to use and give us back hours of time each day.
The winner will be the assistant that knows you best, intuitively solves your problems and becomes the place where you start your work.
