← Back to library

Drive desktop apps when an integration is blocked

Prasid of The Growth CMO kept running into integration walls: an Outlook connector his IT contact would need to approve, a broken messaging link. So he let Claude control his desktop. Over one evening it opened the real apps, wrote emails, texts and a Teams note, waited for his approval, sent them, and placed a draft into his publishing tool.

  • OutcomeGet work done through apps you cannot connect by API, by letting an agent control your desktop and operate them the way you would, with you approving sends.
  • SetupUnder an hour
  • LevelIntermediate
  • CreditPrasid
View original
Works with
Claude Cowork

Published Oct 23, 2026Updated Oct 23, 2026AI dry run on sample input: passed Oct 23, 2026Source verified Oct 23, 2026

When should you use it?

  • An app you need has no approved integration
  • You can let an agent control your desktop
  • You want to approve each send yourself

Not for: Letting it send messages without your approval; Company apps where desktop automation breaks policy

What do you need, and what do you get?

You need

  • An agent with desktop control enabled
  • The desktop apps you want it to operate
  • Clear instructions on what to send and to whom

You get

  • Emails, texts and messages composed in the real apps
  • A human approval before each send
  • Drafts pasted into tools that lack a connector

How does it work, step by step?

  1. Try the integration firstReach for the native connector; use desktop control when it is blocked.
  2. Enable computer useGrant the few permissions needed for the agent to drive the screen.
  3. Spell out recipient and messageName who gets what for every task.
  4. Let it operate the appWatch it open the app, find the thread and compose as a person would.
  5. Approve before it sendsKeep a human gate on every send, then let it hit send.

The prompt

Our own prompt, inspired by Prasid and written to work in Claude Cowork. Values in lime are placeholders.

Prompt0 variables
I need to send a message in [app], which has no approved integration. Using desktop control, open the app, find the conversation with [recipient], and compose this message: [message]. Operate it the way I would. Stop and show me the draft for approval before you send anything. If a step needs a permission I have not granted, pause and tell me rather than working around it.

Inspired by Prasid: I Gave Claude Control Of My Computer for 3 Hours to Help Me Automate My Life. The prompt above is written by playbooks.bot in our own words; nothing from the original is quoted.

What can go wrong?

  • Desktop automation sidesteps approvals your company may require; check policy before using it at work.
  • You still dictate every message, so this removes the typing, not the deciding.

Our take

Prasid's discovery is that an agent can operate software the way a person does, clicking and typing in the actual apps, which gets around missing or blocked integrations. That is a genuine unlock for personal tools, but it deserves a caution the post only hints at. When a company requires IT approval for an email or Teams integration, that requirement is often a security control, and driving the desktop to avoid it sidesteps the policy rather than satisfying it. At work, ask first. Where it is allowed, keep the approval pause before every send, as he did, because a desktop agent can just as easily message the wrong person in a crowded contact list. Spell out recipient and content precisely in each request. Be realistic about the gain, too: you still decide and dictate each message, so the agent removes typing and app-switching rather than thinking. Desktop control is also slower and more fragile than a proper integration, so treat it as a bridge until the sanctioned connection is approved.

Where the time goes back: Work stuck behind missing or unapproved integrations gets unblocked, because the agent operates the real apps instead of waiting for an API.

FAQ

When is this worth it?
When an app has no approved integration and you still need the agent to act in it.
Does it send on its own?
It pauses for approval before sending, so a human stays in the loop on each message.

Source and credit

Inspired by Prasid. We describe the idea in our own words, link to the original, and wrote the prompt ourselves; nothing is quoted.

Inspired by
I Gave Claude Control Of My Computer for 3 Hours to Help Me Automate My Life
Creator
Prasid
Published on
thegrowthcmo.co
Original date
Apr 22, 2026
License
Not openly licensed (the poster keeps all rights), so we quote nothing and link to the original
Originally for
Claude Cowork
AI dry run
Oct 23, 2026: passed. Dry-run tested by our AI test agent on sample input. PASS: Gates held on desk check; no live sends or purchases.
AI dry run on sample data; not run with live connected accounts. The playbooks.bot test agent filled the prompt with synthetic inputs, produced the output and checked it against the page's stated outputs and guardrails.
Verified
Oct 23, 2026: source link resolves to the original, the credited creator published it, our excerpt or prompt matches it, and the license allows our use.

Are you the creator? Claim or correct this page.

More in Operations & Admin

All 203 Operations & Admin playbooks →

Related playbooks

Operations & AdminPrompt✓ AI dry run

Audit how you manage your agent team, not only the agents

Have your chief-of-staff agent review the last 30 days of how you delegated, clarified, interrupted, corrected, assigned, approved and followed up across your agent team, score eight management areas with evidence, list your three biggest weaknesses, and draft a seven-day improvement plan, without changing any bot, routine or instructions until you approve.

Credit: Michael Fenech (@Michael_Fenech_)

Under an hour
Operations & AdminPrompt✓ AI dry run

Brief an agent with nine fields, not a how-to

Before you hand a job to an agent, fill in nine short fields that name the result, the sources it may use, the deadline, what it may do alone, what needs your say-so, what it must never do, how to prove the work is finished, what to do when it gets stuck, and where to put the deliverable, then add a line that tells it to pick the simplest reliable method.

Credit: Michael Fenech (@Michael_Fenech_)

Minutes
Operations & AdminPrompt✓ AI dry run

Brief your agent on the outcome, not every step

Hand a new job to your lead agent by describing the finished result instead of the steps, and have it confirm what success looks like, ask only for context it cannot find itself, pick the simplest dependable approach, state what it will do alone and what needs your approval, and check the result against your goal before calling it done.

Credit: Michael Fenech (@Michael_Fenech_)

Minutes
Operations & AdminPrompt✓ AI dry run

Decide whether an overloaded agent should be split in two

Have your lead agent review two weeks of instructions, routines, task history and corrections for every agent, find where work actually breaks down, separate overload from fixable causes like vague instructions or missing access, give each agent a keep, simplify or split verdict, and propose one measured trial before anything changes.

Credit: Michael Fenech (@Michael_Fenech_)

Under an hour
Operations & AdminPrompt✓ AI dry run

Find routines your agents run twice and merge them

Have one agent collect every scheduled routine across your agent team, compare the saved instructions of any that look alike, flag pairs that read identical sources or run too close together, and hand you a short keep, merge or cut table with the expected saving, without changing anything until you say yes.

Credit: Michael Fenech (@Michael_Fenech_)

Minutes