How Claude Apps differ from Claude Code, Cowork, and the raw API
| Surface | Who it's for | How it's built | Where it runs |
|---|---|---|---|
| Claude Apps (Artifacts) | Anyone with a Claude.ai account, coder or not | Natural-language conversation in the Artifacts space; "no coding required" | Anthropic's own infrastructure; a published app's usage is billed against each viewer's own Claude subscription |
| Claude Code | Developers working in a terminal/IDE on a real codebase | CLI-driven agentic coding sessions against local files, git, and shell | The developer's own machine or a Managed Agents cloud sandbox |
| Claude Cowork | Operators delegating multi-step work across tools | A managed-agents session with Contract/Task/Memory primitives, gated approvals, and MCP connectors | Cloud sandbox / managed-agents infrastructure |
| Claude Agent SDK / raw API | Developers embedding Claude in their own application | Python/TypeScript SDK code calling the same agent loop and context management that power Claude Code, or direct Messages API calls | The developer's own infrastructure, under the developer's own API billing |
Per the Agent SDK's own docs: "The Agent SDK is a collection of tools that helps developers build powerful agents on top of Claude Code... programmable in Python and TypeScript," giving "the same tools, agent loop, and context management that power Claude Code." One concise framing found this session, from a third-party comparison rather than Anthropic directly (flagged as such, not presented as an Anthropic quote): "Claude Code is the finished product you use for interactive coding; the Claude Agent SDK is the extracted engine you embed in custom applications." Claude Apps sit outside that developer-tooling lineage entirely -- they are a consumer/prosumer no-code surface, not something built with the Agent SDK or the raw Messages API by the app's creator. The one place the two worlds meet: an AI-powered Claude App itself calls Claude's intelligence "through a text-based API" at runtime, but that call is made by Anthropic's own hosted artifact runtime on the creator's behalf -- the app's creator writes no SDK or API-key integration code themselves.
Honest gap
No single Anthropic page directly enumerates all four surfaces side by side; this comparison synthesizes claude.com/blog/build-artifacts (Claude Apps) against code.claude.com/docs/en/agent-sdk/overview (Agent SDK / Claude Code lineage) and this repo's own existing grounding for Cowork (mirrored under docs/docs/claude.com/docs/claude-tag/ and docs/docs/cowork/). Anyone auditing this page should treat the table as this site's own synthesis, not a verbatim Anthropic comparison chart.
Sources: code.claude.com/docs/en/agent-sdk/overview, claude.com/blog/build-artifacts. Live-fetched/searched 2026-07-02.