A practical guide to Claude Chat, Artifacts, Cowork,
Skills, Connectors, Agents, n8n, and the API
The further right → the more powerful, the more AI can do for you
Browser-based conversation at claude.ai. You type, Claude responds. Supports connectors if set up — so Claude can also pull from Slack, Drive, or Calendar directly.
Paste in a product update → ask for three email subject line variations for different audience segments. Or, with a Slack connector, ask Claude to summarize #product-updates this week — without copying anything yourself.
Artifacts don't auto-save. When done refining, copy it out to Confluence, a doc, or wherever it belongs. Don't assume Claude is saving your work.
A desktop app that runs on your computer with access to your local files and folders. Point it at a folder, describe a task — Claude works through it without you opening each file.
Chat works from what you paste in. Cowork accesses your actual machine. No copying, no pasting — just point and describe.
Cowork handles the input (your files). Artifacts handle the output (the rendered deliverable you refine). They work together in one session.
Pre-packaged instructions that tell Claude how to do a specific task — the process, format, and rules to follow. Anthropic ships some out of the box (PowerPoint, Excel, Word, PDF). Teams can build custom ones.
Without a skill, Claude improvises — good, but inconsistent. With a skill, it follows the same tested process every time. Less cleanup, more predictability.
Skill = the playbook Claude follows · Artifact = the thing Claude produces using that playbook. One is input; one is output.
// same concept, different name depending on the tool
You can install connectors yourself — no ops needed. In the Claude desktop app go to Settings → Extensions and drag in a .MCPB or .DXT file, or browse the gallery. Each extension needs an API key from the service it connects to. Most are read-only — always check the docs before assuming it can write or act.
Multi-step tasks you kick off manually, within a session. Great for one-off complex jobs — research, write, save, send. You trigger it. It runs once.
Anything that needs a schedule, an event trigger, or has to run without you initiating it. More robust, persistent, built for production.
Claude is a product you interact with. An agent is a concept — an AI model running steps autonomously inside a workflow. In n8n, you use an AI agent node powered by an Anthropic model, not Claude as a product. You can't use Claude directly in n8n.
Cloud-based, company access provided. AI agent nodes call Anthropic models directly — no chat interface, no Claude product. The model is just one step in a larger workflow.
Genie is a Databox feature built on our own agentic platform — separate from n8n. You can trigger it from n8n workflows via the Databox MCP, but Genie itself is not an n8n workflow.
Always test without activating first. If a workflow touches a live system (Slack, Asana, email, database) → get ops sign-off before activating.
The API is the developer interface — how engineers connect to Anthropic models programmatically to build custom tools, products, and integrations. It's the lowest level of access, and the most flexible.
You won't use the API directly — that's engineering territory. But knowing it exists means you can propose ideas. "Could we build something that automatically does X?" is a valuable question. Bring it to ops — the API is probably how they'd answer it.
| Situation | Reach for |
|---|---|
| One-off task, everything fits in a message | 💬 Claude Chat |
| You want a deliverable to refine and copy out | 📄 Artifact |
| Task involves your local files or folders | 🖥️ Cowork |
| Recurring task needing consistent structure | 🧩 Skill |
| Need Claude to read your Slack, Calendar, Asana | 🔌 Connector |
| High-volume, multi-step, or scheduled task | ⚙️ Agent / n8n |
| Build something custom into a product or system | 👩💻 API → talk to engineering |
// the more Claude does for you → the further right on the spectrum
A plugin is a pre-packaged bundle that combines connectors, skills, and tools in one installable file. Instead of setting things up one by one, you install a plugin and get everything configured at once.
Think of it like installing an app — but for Claude's capabilities.
Plugins lower the barrier. You don't need to know how MCP works or how to write a skill — you just install the plugin and start using it. As we build more internal plugins, this is how the team will get access to new AI capabilities.
The content, the script, the voiceover — all produced with AI.
The only manual step was assembling it in a video editor.
// practice what we preach