AI Tip — August 2026

A normal conversation with Claude is a back-and-forth — you say something, it replies, you correct it, it adjusts. Cowork isn’t quite that. You hand it a task and it goes off and does a run of work on its own — reading files, searching, writing, sometimes publishing — before it comes back to you. On the iPhone app especially, you might fire off a prompt between other things and not be back at your screen for an hour. That changes what a good prompt needs to contain. It has to do more of the thinking for you up front, because you’re not going to be there to steer it turn by turn.

This is what I’ve learned works, plus the groundwork worth doing before you type anything at all.

Three things to have in place before you start

1. Put your source material somewhere Cowork can actually reach

On the iPhone app, Cowork can’t reach into your phone’s photo roll or your Files app on its own, and there’s no cable running to your laptop. What it can reach is anything already living in a connected cloud service — Google Drive, Gmail, your calendar — or anything you attach directly to the chat. If the thing you need it to work from is “a document somewhere on my phone,” stop and get it into Drive first, or attach it there and then. Otherwise the task stalls on the one thing it can’t do: go looking through your phone for you.

2. Connect the accounts the task actually needs

Gmail, Google Calendar, Google Drive, WordPress, Craft — each one is a one-off setup, and then it’s just there for every future task. Worth doing this in a quiet moment rather than mid-task, because a task that needs your calendar and finds it isn’t connected yet either stalls or, worse, guesses.

3. If it’s a recurring subject, set up a Project first

A one-off task doesn’t need this. But if you’re going to be asking for the same kind of thing repeatedly — blog posts, medical summaries, household guides — a Project with your instructions and reference documents already sitting in it means you’re not re-explaining your setup every single time. You write the context once. After that, every prompt in that Project starts from what it already knows.

The four things a good Cowork prompt should say

What you actually want, specifically. “Write about AI” gets you something generic. “Write a blog post evaluating five ways X could help Y, with pros, cons and a verdict for each” gets you this post.

Where the material should come from. Say if it should use a specific file, a folder in Drive, information already in the conversation, or go and research it fresh. Don’t leave this implicit — it’s the single biggest cause of Cowork doing more (or less) digging than you wanted.

Any constraints that matter. Length, tone, format, what to avoid, who it’s for. If you have a house style — plain language, no bullet points, whatever it is — say so, or better, have it saved somewhere Cowork already reads.

Where the result should land. A file back to you, a WordPress draft, a published post, a saved copy somewhere specific. Cowork will make a reasonable guess if you don’t say, but your reasonable guess and its reasonable guess aren’t always the same thing.

Weak prompt versus strong prompt

Weak: “Write something about how AI can help disabled people.”

Strong: “Write a blog post listing eight to ten Cowork tasks that could help a disabled person, evaluating each with pros, cons and a worth-it verdict. Match the tone of my existing posts. Save a copy to Craft in the usual folder and publish it to my WordPress blog with relevant tags.”

The second one is longer, but it’s one exchange instead of four rounds of “what did you mean by that.”

Say if you want it to check with you first

Because Cowork can act on its own — send things, publish things, delete things — it’s worth being explicit when you want a checkpoint before anything goes out into the world. Something like “draft it and show me before you publish” costs you one extra reply, but it means nothing goes live that you haven’t actually seen. Cowork will generally ask before anything irreversible-feeling on its own, but don’t rely on it guessing right about what counts as irreversible to you.

The honest summary

A good Cowork prompt on the phone does the job a good written brief does anywhere else: it says what you want, where the material is, what the boundaries are, and what happens when it’s done. The setup — Drive, connectors, a Project if you need one — is a one-time cost. Do it once and every prompt after that gets shorter, not longer, because Cowork already knows where to look.

Written in conversation with Claude.

This blog is written with the assistance of Claude. That’s not a footnote — it’s rather the point.