Claude’s Hidden Modes: Why One Chatbot Is Actually Five Tools in Disguise
## The Illusion of a Single Interface The screen looks the same: a text box, a send button, and a scrolling transcript. But for users of Claude, Anthropic’s flagship AI assistant, that familiar interface is a Trojan horse. Beneath the surface lie five distinct tools, each designed for a specific cognitive task—design, coding, memory, or administrative work. Most users never discover them. Artemiy Miller, a Russian tech educator with a following in AI circles, recently broke down the five modes in an Instagram reel. His observation is simple but jarring: *Claude isn’t one tool. It’s five.* The default chat interface, where users ask questions or brainstorm ideas, is just the first. The other four—Design, Code, Projects, and Cowork—are specialized agents that operate under the same brand but serve entirely different functions. Yet they’re not advertised on Anthropic’s homepage, nor are they mentioned in the company’s official documentation. They exist as unspoken capabilities, accessible only to those who know the right commands or stumble upon them in niche communities. This isn’t just a quirk of product design. It’s a microcosm of how AI is evolving: from a single, general-purpose assistant into a network of specialized agents, each optimized for a narrow but high-value task. The shift mirrors the broader trajectory of software itself—from monolithic applications (like early Microsoft Office) to modular, task-specific tools (like Figma for design or GitHub Copilot for coding). But with AI, the transition is happening faster, and the boundaries are blurrier. Users who assume Claude is just a chatbot are missing out on 80% of its utility. ## The Five Faces of Claude Miller’s breakdown of Claude’s modes is worth reproducing in full, because it reveals how each tool is tailored to a distinct workflow: 1. **Chat** – The default mode. Think aloud, brainstorm, or ask questions. This is where most users stop. 2. **Design** – A virtual designer. Upload a brand book or provide a URL, and it generates landing pages, visual assets, or marketing materials without a formal brief. No calls, no revisions, no human designer. 3. **Code** – A development team in a box. Describe what you need in plain Russian (or English), and it outputs functional code for websites, apps, or automations. The “vibe coding” approach—where intent is translated directly into code—eliminates the need for technical specifications. 4. **Projects** – A second brain. Upload past chats, documents, or datasets, and Claude retains the context. Need to recall a conversation from weeks ago or build a new task based on old data? Projects acts as a memory layer, often paired with tools like Obsidian for local file management. 5. **Cowork** – A virtual assistant for administrative work. Handles routine tasks like reports, audits, or calendar management. It operates on local documents and integrates with external tools (e.g., Google Calendar), functioning as a persistent, context-aware aide. The list is striking not just for its breadth, but for its specificity. Each mode isn’t just a tweaked version of the chat interface—it’s a fundamentally different interaction model. Design, for example, doesn’t just generate text descriptions of visuals; it produces actual design assets. Code doesn’t just explain how to write a function; it writes the function for you. Projects doesn’t just summarize past conversations; it *remembers* them, allowing for multi-step reasoning across long time horizons. This raises an obvious question: Why aren’t these modes more visible? Anthropic’s decision to tuck them away suggests a deliberate strategy. The company may be avoiding the complexity of explaining five distinct tools to new users, opting instead for a gradual onboarding process. Or it may be testing these features in stealth, refining them based on usage data before a wider rollout. Either way, the approach creates a two-tiered user base: those who treat Claude as a chatbot, and those who wield it as a Swiss Army knife of AI agents. ## The Tradeoffs of Specialization The existence of these modes reflects a core tension in AI development: *generalization vs. specialization*. Large language models like Claude are trained on vast, diverse datasets, giving them the ability to handle a wide range of tasks. But raw generality comes at a cost. A model that can write poetry, debug code, and design a logo is less likely to excel at any one of those tasks compared to a model fine-tuned for a specific domain. Anthropic’s solution is to offer both. The default chat mode is the generalist, while the other four modes are specialized agents that leverage the same underlying model but with different prompts, context windows, or tool integrations. This approach mirrors the rise of *mixture-of-experts* (MoE) architectures in AI, where a single model is composed of multiple sub-models, each optimized for a specific type of input. In Claude’s case, the “experts” aren’t just internal—they’re exposed to the user as distinct modes. The tradeoff is complexity. For power users, the ability to switch between modes is a force multiplier. A startup founder, for example, could use Design to create a landing page, Code to build a prototype, and Projects to manage the project’s context—all within the same ecosystem. But for casual users, the learning curve is steep. The Instagram comments under Miller’s post reveal the divide: some users are thrilled by the discovery, while others dismiss it as overhyped (“Show me a real result, not just ‘phenomenal’ claims”). This dynamic isn’t new. It echoes the early days of cloud computing, when AWS offered dozens of services that most users never touched. Or the rise of Photoshop, where 90% of users relied on 10% of the features. The difference with AI is that the tools are more abstract. A chatbot doesn’t come with a toolbar or a manual; its capabilities are hidden behind natural language prompts. This makes discovery harder, but it also makes the payoff greater for those who invest the time to learn. ## Who Benefits—and Who Loses The primary beneficiaries of Claude’s hidden modes are power users: developers, designers, entrepreneurs, and knowledge workers who can integrate AI into their daily workflows. For them, the ability to offload design, coding, or administrative tasks to specialized agents is a game-changer. It’s not just about saving time; it’s about reallocating cognitive bandwidth. A founder who no longer needs to hire a designer for a landing page or a developer for a simple automation can focus on higher-leverage work. But there’s a catch. These modes aren’t just about efficiency—they’re about *skill substitution*. Claude Design, for example, doesn’t just assist designers; it replaces the need for a designer entirely, at least for basic tasks. The same goes for Claude Code, which can turn a non-technical founder into a solo developer. This democratization of expertise is a double-edged sword. On one hand, it lowers the barrier to entry for creative and technical work. On the other, it devalues the skills of professionals who once filled those roles. The losers, then, are likely to be freelancers and agencies that rely on routine work. A junior designer who charges $50/hour to create landing pages is now competing with an AI that can generate one in seconds. A developer who writes boilerplate code is similarly at risk. This isn’t hypothetical; it’s already happening with tools like GitHub Copilot, which has been shown to reduce the time developers spend on repetitive coding tasks by up to 55%. Claude’s modes simply extend this trend to new domains. There’s also a risk for Anthropic itself. By hiding these modes, the company is betting that power users will discover and evangelize them, creating a grassroots adoption curve. But if the features remain obscure, they may never gain enough traction to justify their development. The alternative—aggressively marketing them—risks overwhelming casual users, who might abandon the product entirely if it feels too complex. ## The Bigger Picture: AI as a Network of Agents Claude’s five modes are a preview of where AI is headed. The future isn’t a single, all-knowing assistant, but a network of specialized agents that collaborate to solve complex tasks. This vision aligns with the emerging paradigm of *multi-agent systems*, where multiple AI models work together, each handling a specific part of a workflow. For example, one agent might research a topic, another might draft a report, and a third might edit it for clarity—all without human intervention. Anthropic’s approach is a step toward this future, but it’s still limited. The modes are siloed; they don’t yet collaborate with each other or with external tools in a seamless way. Projects, for instance, can remember context, but it’s not clear whether Design or Code can access that context automatically. The next frontier will be breaking down these silos, allowing the agents to work together as a unified system. For now, though, the most striking insight from Claude’s hidden modes is how much of AI’s potential remains untapped. Most users interact with these tools in the simplest way possible, unaware of the depth beneath the surface. The challenge for companies like Anthropic isn’t just building more powerful AI—it’s designing interfaces that reveal that power without overwhelming the user. Until then, the divide between casual users and power users will only grow wider, and the true capabilities of AI will remain hidden in plain sight.