Mapping the Claude Ecosystem: Four Products, One

Most people who use Claude use it the same way: open a tab, type a question, get an answer, close the tab. That's a perfectly reasonable way to interact with it. It's also, according to a recent Futurepedia breakdown, roughly equivalent to buying a Swiss Army knife and only ever using the toothpick.

The Futurepedia video lays out what Anthropic has actually built across the Claude platform—four distinct products, a cross-cutting layer of customization features, a memory and context system, and a set of autonomous capabilities that don't require a human in the loop at all. Whether that picture excites or unsettles you probably says something about your threat model. Either way, the map is worth understanding.

Four Products, Not One

The video's core framework is clean: Chat to think, Co-work to delegate, Code to build, Design to create. Each product shares a surface-level resemblance—you type, Claude responds—but the underlying model of what Claude is doing differs significantly.

Chat is conversational and synchronous. You're in the loop for every exchange. It's the product that most resembles what people mean when they say "AI chatbot," and it's useful precisely because of that interactivity—working through a problem, drafting something iteratively, doing research that requires follow-up questions.

Co-work operates differently. You assign a task, and Claude executes it against your actual file system and connected tools—email, calendar, documents—without your step-by-step involvement. As the video describes it: "You hand it a task, you walk away, and you can come back to finished work. You're not in the loop for every little step, you're just reviewing the result." The example given—Claude finding a meeting on your calendar, reading your notes, building the presentation, and drafting the follow-up email, all without you touching it—is the kind of workflow that sounds either liberating or alarming depending on how much you think about what "access to your email" actually means. Our earlier reporting on Co-work's desktop automation and its implications for non-technical users covers that tension in more depth.

Claude Code extends the no-coding-required promise further: describe an app, a Chrome extension, a dashboard in plain English, and the product writes functional code and ships it. "A complete beginner can sip their morning coffee while describing a game or an internal tool or dashboard, then have something working before lunch," the video suggests. That framing—"so now you are a developer"—is bold, and it's worth sitting with. There's a difference between generating working code and understanding what the code does, a distinction that matters considerably when that code is touching your data or running automations on your behalf.

Design, the newest of the four, handles prototypes, slide decks, landing pages, and animations from prompts or existing brand assets. It's positioned as the front end of a pipeline: design something, then hand it to Code to ship. The Figma integration that enables this kind of handoff is part of a broader connector ecosystem the video describes next.

The Universal Layer: Skills, Connectors, Plugins

Sitting underneath all four products is what the video calls a "universal layer"—features that extend and customize Claude's behavior across Chat, Co-work, and Code.

Skills are essentially reusable instruction sets. You define how you want a recurring task executed—your writing voice, your formatting preferences, your brand guidelines—and Claude stores that as a skill it calls automatically whenever that task comes up. "What this really is underneath is just a folder of instructions and guidelines that Claude reads first," the video explains. That's a useful demystification: it's not magic, it's prompt engineering packaged for reuse. The Claude Skills feature, which allows users to build custom automations through conversation, is worth understanding in this context—particularly the questions about where those instructions live and who can access them.

Connectors link Claude to external tools—Gmail, Slack, Google Drive, Asana, and hundreds more. Permissions are user-controlled, which is the right design. What the video doesn't dwell on is the question of what "user-controlled" means in practice when most users don't read permissions dialogs carefully, and when the scope of "Claude can take actions through that tool" is fairly broad.

Plugins bundle skills and connectors together around specific use cases—a Figma plugin for design workflows, a legal plugin for drafting briefs and managing research. Anthropic runs a plugin marketplace. The convenience argument is strong. The third-party trust question is real, though the video doesn't raise it.

Memory Without Amnesia

The memory and context layer is where the architecture gets genuinely interesting—and where the distinctions matter most.

Projects are containers for related conversations. Inside a project, custom instructions persist, uploaded files provide context, and memory refreshes automatically across sessions. A well-configured project means Claude always knows your voice, your clients, your past results—you don't re-explain yourself every time. Projects in Chat and Co-work are separate and don't automatically sync, though you can import from one to the other.

Account-level memory applies only to Chat, not across all products. It learns your preferences, working style, and context over time—a persistent model of you as a user.

Claude.md is Code's equivalent: a markdown file generated when you run /init in a project folder. It holds everything Claude needs to understand the project—context, conventions, architecture decisions—and updates as work progresses. Some users create these manually for Co-work projects too.

These are meaningfully different memory models, which creates a fragmentation that isn't entirely resolved. Your Chat history knows you; your Code projects are briefed per-folder; Co-work sits somewhere between. The practical implication is that the "Claude who knows you" isn't a unified entity—it's several different context windows with different persistence rules.

The Autonomous Layer

This is where the video gets into territory that deserves more scrutiny than a 12-minute overview can provide.

Computer use gives Claude direct control of your desktop—opening apps, navigating browsers, filling out forms, moving files. The video acknowledges it's early-stage: "It's still early, so you will run into some issues, but it's a glimpse at where these tools are headed." That's honest. It's also worth noting that "AI agent with full desktop control" is a meaningful capability expansion, and the error modes are different from a chatbot getting a fact wrong.

Scheduled tasks in Co-work let you configure recurring automations—a morning briefing that scans your calendar and email, a weekly competitive research report, an end-of-day summary sent to your team. These run on a schedule without manual triggers.

Claude Code Routines go further: automated code execution triggered by schedules, GitHub events, or API calls, running without requiring your computer to be on. That's a server-side autonomous agent, which is a different category of tool than a productivity assistant.

Dispatch lets you monitor and redirect a Co-work session from your phone. It's the human-oversight feature in a set of tools that otherwise emphasizes hands-off operation.

The pattern across all of these is autonomy: Claude doing things without waiting for you. That's the pitch. The question every user should ask before enabling these features is not "what can this do?" but "what happens when it does the wrong thing, and how do I know?"

What This Architecture Actually Is

Step back and the picture that emerges is less "AI chatbot with extras" and more a modular automation platform with a conversational interface. Chat is the part most people know. Co-work, Code, and Design are increasingly agentic tools that operate against real systems—your file system, your email, your calendar, your code repositories.

The video frames this as progressive empowerment: you can engage at whatever depth suits you, from basic chat to fully automated pipelines. That framing is accurate. It's also worth noting that the complexity is real—the video exists because the ecosystem has grown to the point where a 12-minute explainer is necessary just to map the terrain.

"The more powerful it gets, the more overwhelming it becomes to understand and use it all," the Futurepedia presenter says at the top. That's not a critique—it's an honest acknowledgment that capability and complexity are correlated. The question for users is whether they want to engage with the full depth of what's been built, or whether they're comfortable using the toothpick while the rest of the knife sits folded.

Both are legitimate choices. The prerequisite is knowing the knife has other blades.