The thinking tool. If the work needs care, hand it to Claude.
For UK operators who actually read what the model produces, Claude is the default. The writing has a steadier register than ChatGPT, Projects plus Artifacts turn it into a proper desk for proposal work, and Claude Code has quietly become the centre of gravity for anyone building internal tools. The price of admission, if you go beyond the free tier, is that you need to learn Projects, Skills, and at least one MCP connector; without those it is just a smarter chatbot.
Reach for Claude when the work is reading-and-writing heavy and quality matters: contracts, proposals, board papers, code reviews, long-form research that the recipient will actually read. Reach for it as the default model whenever Projects and Artifacts let you load the full context once and iterate against it. For any UK SME that drafts client-facing documents weekly, Claude Pro or Max plus a couple of good MCP connectors (Drive, Gmail) is the single best AI investment available.
Do not reach for Claude when you need native image generation (use ChatGPT's DALL-E or Midjourney directly). Do not reach for it for full voice conversations on a phone (ChatGPT or Gemini are ahead). Do not reach for it when the job is genuinely real-time web research with cited sources at the URL level (Perplexity is sharper, ChatGPT search is fine). Do not pay for Max if you only use the chat for a few prompts a day; Pro is enough.
Long-form reasoning, document work, and developer-grade agentic execution. Claude is the model you trust with the contract, the board pack, the codebase migration; it is not the model you reach for to generate a hero image or scrape Twitter in real time.
All three are Claude. What differs is where the conversation happens, what Claude can touch, and how the output persists. Pick the one that matches the SHAPE of the work, not the difficulty of the task.
| If the output needs to be… | Reach for |
|---|---|
| A response you read in the chat (draft, summary, answer) | claude.ai |
| A persistent dashboard that refreshes against live data | Cowork (Live Artifacts tab) |
| A file (or files) on your local disk | Cowork (agentic tasks) |
| A scheduled recurring job (e.g. every Monday 07:00) | Cowork (scheduled tasks) |
| Code in a repo, agent infrastructure, custom skills you ship | Claude Code |
| claude.ai (browser + app) | Claude Cowork (desktop) | Claude Code (CLI) | |
|---|---|---|---|
| Audience | Everyone | Operators, knowledge workers | Developers, agent builders |
| Where it runs | Browser, mobile, desktop | Desktop only (macOS + Windows) | Local CLI on macOS / Linux / Windows |
| Output type | Chat responses + inline Artifacts | Live Artifacts (persistent tab) + files on disk | Code, agent runs, structured output |
| Persistence | Tied to chat (saveable to Project) | Live Artifacts tab is permanent | Files persist in your repo |
| File access | None (only via upload to a Project) | Yes — read, edit, create local files | Yes — same as Cowork |
| Multi-step execution | One prompt at a time | Agentic plan + approval flow | Agentic; usually scripted via skills |
| Scheduled tasks | No | Yes (Cowork scheduled tasks) | Yes (via cron + claude -p) |
| MCP connectors | Yes (Gmail, Drive, Calendar, etc.) | Yes (same connectors) |
Each sub-product has its own focused page. Tap any card to open the full breakdown.
Reach for this when the deliverable needs to read like a competent senior wrote it.
View details →Reach for this when you would otherwise hire a junior developer for two weeks of glue work.
View details →Reach for this when the work lives across local files, email, and Drive and you do not want to script it yourself.
View details →Reach for this when you are wiring Claude into your own product or back-office and want production-grade behaviour.
View details →Reach for this when you are on site or in a car and need the same Claude you use at your desk.
View details →Useful connectors worth knowing for this tool. Ranked by everyday usefulness.
Gives Claude direct read and write access to nominated local folders. The non-negotiable MCP for anyone using Claude as a desk agent over real client files.
Read repos, open issues, comment on PRs, and run Claude reviews on diffs without leaving the chat. Essential for any team using Claude Code alongside GitHub.
OAuth connectors in Settings that let Claude read attachments and threads. The right way to give Claude inbox context without exporting to a Project.
Reads channel history, summarises threads, and posts messages. Pairs neatly with Cowork scheduled tasks for a weekly digest.
Injects version-correct library documentation into Claude Code sessions. Stops the model hallucinating against a 2024 version of your framework.
Drives the desktop and browser directly. Useful for legacy web apps with no API; treat as a junior assistant, not a fire-and-forget agent.
Pre-built skill packs for vendor contract review, marketing outreach, financial modelling. The fastest way for a non-developer to get value from Claude beyond chat.
Gives Claude a proper page-fetcher and crawler. Closes the gap with ChatGPT and Perplexity on web research inside a Project.
| Yes (same connectors) |
| Best learning surface for first-timers | Yes | Once chat is comfortable | Last; assumes CLI fluency |
A non-technical operator does 90% of their AI work in claude.ai. The browser is the iteration loop. Projects hold context that travels across conversations. Custom Instructions hold voice rules so output is consistent. Artifacts emerge inline when you ask for "an interactive…" or "an editable…".
A senior operator moves to Cowork when the work is recurring or file-shaped. A weekly digest that refreshes from Gmail goes here. A folder of contractor RAMS that needs scanning goes here. A dashboard you reopen every Monday morning goes here.
An engineer reaches for Claude Code when the work is code, skills, or agent infrastructure. Skills written here ship via .claude/skills/ and become callable from anywhere. Agent runs orchestrate multiple tools and produce structured output for downstream pipelines.
Most teams use all three. The skill is knowing which to start in for a given task. The wrong starting point is reversible but costs minutes.
These four often confuse people. Here is how they relate.
The decision rule: a skill is a graduated prompt. A Project is a graduated chat. Artifacts are what the model produces when you ask for interactivity. Custom Instructions are how you bake your voice into every output.
Cowork has three modes of work:
You often combine them. A Live Artifact you maintain manually + a scheduled task that refreshes it every Monday morning. An agentic task that produces files + a Live Artifact that lets you browse those files interactively.
May 2026. Anthropic ships new features fast; if a date below disagrees with the current product, the product wins.