Why every AI company wants to become your work assistant

The article argues that competition in artificial intelligence has shifted focus. Two years ago the debate revolved around models: which chatbot was smarter, who dominated the benchmarks and who had the largest context window (the piece recalls how Anthropic expanded Claude from 9,000 to 100,000 tokens in 2023, and how Google responded in February 2024 with a one-million-token window for Gemini 1.5). That race is still alive, but according to the author it has already been displaced in importance by another: becoming 'the assistant' people use to start their workday.

The central thesis is that the value no longer lies in the model alone, but in accumulated context. An assistant that knows your files is useful; one that also knows your colleagues, your clients, your preferences, your projects and your way of working is far more useful and, above all, far harder to abandon. That accumulation of data and habits acts as an exit barrier: the longer an assistant has spent learning how you work, the more painful it is to migrate to another.

The text describes a change in behavior: previously, to send an email you opened Gmail, to find a document you went into Drive, to update a client you used Salesforce. With an assistant, by contrast, the user simply asks for the task ('find the latest figures, update the forecast and let Sarah know') and it is the assistant itself that decides which applications to use. Software platforms remain essential infrastructure, but they cease to be the starting point of daily work, and that 'gateway' role is the one every major AI company is fighting to occupy.

The article reviews six different bets, each starting from a different advantage:

OpenAI leans on ChatGPT as a mass-market destination and is broadening its offering with ChatGPT Work, Workspace Agents and, more recently, Dots, always-on agents with their own computer in the cloud, capable of learning preferences and connecting to thousands of applications via plugins. Its advantage is not depending on any particular suite and being able to present itself as a cross-cutting layer of intelligence; its challenge is earning the access to context and the large enterprise contracts that Microsoft and Google already have.

Google plays the opposite card: it already owns much of the working (and personal) life of millions of users in Gmail, Calendar, Docs and Drive. Gemini increasingly acts across all those applications, and Google positions Gemini Enterprise as the 'gateway' for corporate AI, alongside Gemini Spark, a personal agent available 24/7 and built into Workspace. Its advantage is the context it already has and the absence of switching friction; its challenge is getting Gemini to be perceived as a coherent assistant rather than as AI features added to existing products.

Meta is making the boldest play: Muse, a personal agent that runs on its own secure virtual computer, can work across connected applications, remember user information and act on the user's behalf. Meta is bringing Muse to its AI glasses, which could give it access to what the person sees while moving through the physical world, and it has launched Muse for Small Business, with connections to tools such as Shopify, QuickBooks, Slack, Stripe, Canva, Notion and Facebook and Instagram's own business products, as well as a broader Enterprise Platform (Muse, Meta Business Agent, Muse Code). While Microsoft and Google start from the work environment, Meta proposes starting from the user's physical body. Its advantage is massive distribution and a hardware route that no traditional enterprise software competitor can easily replicate; its challenge is trust, given the company's reputational track record on privacy, something especially sensitive when asking for access to such personal information.

Microsoft starts from its dominant position in the enterprise: Copilot sits alongside the stack of tools that millions of companies have relied on for decades, and Microsoft increasingly treats agents as a corporate resource that must be deployed, measured, governed and paid for, with new long-running agentic work capabilities, model choice and centralized usage and spending controls. Its advantage is that it needs no introduction to any CIO; its challenge is that the very same corporate robustness can feel heavy for individual users or small teams who prefer faster, simpler and cheaper tools.

Anthropic is betting on advanced technical knowledge work: Claude Code has established itself as a serious tool for developers, and Cowork extends that same agentic model to broader file-management tasks and multi-step work across applications. The article notes that Anthropic has made notable inroads in large companies, citing PwC — described as the world's second-largest professional services firm — as a client deploying Cowork and Claude Code across its teams. Its advantage is the credibility built around Claude as a tool for serious work; its challenge is competing head-on with giants such as Microsoft and Google, which already control enormous enterprise and consumer contracts.

Grok Bot, from SpaceXAI, presents itself as the simplest option: you create a 'Bot', assign it a task, and it works with a persistent cloud computer, memory, files, browser sessions and its own tools, able to keep working with the laptop closed, to learn workflows by watching the user and to coordinate with other Bots. The author admits to being a daily user of this product and values the fact that, although it occasionally gets stuck, it always finds a clear way to solve the problem permanently, without hitting the same obstacle over and over. Its advantage is a very intuitive mental model — delegating work to a teammate rather than building an automation; its challenge is that, being newer than the big enterprise platforms, it still has to prove that this simplicity scales in large organizations, although it has already begun to incorporate enterprise governance features and shared team Bots.

The article closes with a reflection on what it calls 'the new platform war': far from destroying software platform lock-in, AI could be creating an even stickier layer on top of the current technology stack. These companies are competing not only to build the smartest assistant, but to become the layer where work actually happens, which means distribution, a direct customer relationship and recurring revenue. The economic reasoning is simple: a user may ask a chatbot five questions a day, but a genuinely useful assistant can take part in dozens or hundreds of tasks — sorting email, tracking news, reformatting content, updating documentation — and every new responsibility it takes on makes it more valuable both to the person using it and, crucially, to the company building it. The author's conclusion is that most users do not care whether one model scores two points higher than another on a benchmark, but whether the product is easy to use and gives them back hours of their day; that is why, he says, the winner will be the assistant that knows the user best, solves their problems intuitively and becomes the place where their work begins.