AI Coaching · Part 2

Claude as a coworker,
not a search box

A partner you hand real work to — and correct — rather than a machine you query.

What "coworking with Claude" means

Delegate the work, not just the question.

A search box answers a question and forgets you. A coworker takes on a piece of real work — drafting a proposal, summarising a long report, analysing a spreadsheet, translating a document, planning a project — and stays in the conversation while you shape the result. That is what working with Claude looks like: less “look this up,” more “help me get this done.”

It is a natural step up from the foundation. The three rules still hold — check every output, keep confidential data in the sanctioned workspace, own the final — but the unit of work gets bigger. You stop asking for facts and start delegating tasks.

Two features that make it a coworker

Durable context, and real deliverables.

Projects

A Project lets you give Claude durable context — documents, reference material, and standing instructions it keeps across every conversation inside it. It is the reusable-helper idea from the foundation, grown up: instead of re-briefing from scratch each time, you set up a space that already knows your recurring work, and simply pick up where you left off.

Artifacts

Artifacts are real, workable deliverables Claude produces alongside the conversation — a document, a table, a simple app, a page — that you can see, iterate on, and refine live. Rather than copying text out of a chat, you watch the deliverable take shape and adjust it in place until it is right.

Good cowork patterns

How the best work actually happens.

  • Bring the source material. Paste the report, the thread, the numbers. Claude works far better grounded in your real documents than guessing from general knowledge.
  • Ask for a draft, then correct. Treat the first output as a starting point, not a verdict. The value is in the second and third pass, where you steer it.
  • Ask it to show its reasoning and flag uncertainty. “Walk me through how you got there, and tell me what you are unsure about.” This is how you catch the confident-but-wrong answer.
  • Iterate rather than one-shot. The best results come from a short back-and-forth — refine, narrow, adjust — not from one perfect prompt.
When to reach for Claude

Bigger, longer, and built on your documents.

Reach for the coworker when
  • The work is longer, analytical, or multi-step
  • It needs your own documents as context
  • You will reuse the setup again and again
  • You want a real deliverable, not just an answer
A quick chat prompt is fine when
  • You need one fast fact or a short rewrite
  • There is no document to work from
  • It is a throwaway you will not revisit

A more capable coworker still needs the same discipline.

Nothing about a stronger partner changes the rules. Check every output — the better the draft, the easier it is to stop reading, which is exactly when errors slip through. Keep confidential data inside the sanctioned workspace. And remember who owns the final: the coworker drafts, reasons, and iterates, but the human signs off. Capability raises the ceiling; it does not remove the check.

Next in the curriculum

A coworker drafts and reasons in conversation. The next leap is AI that operates — working directly on your files and systems. Continue to Claude Code, revisit the foundation, or return to the curriculum hub.

Put a coworker on the real work.

Coaching that moves teams from asking questions to delegating outcomes.

Start a Conversation