Windows only
Desktop flows and computer use run on Windows machines. Mac and Linux desks are not covered.
a Microsoft platform for cloud flows and desktop automation, paired with Copilot Studio for building AI agents across Microsoft 365
Microsoft offers three ways to automate, each with its own meter.
Steps
Connector work across Outlook, Teams, SharePoint, Dataverse, and approvals. Premium is $15 per user a month.
Drive Windows apps through UI selectors (UIA, MSAA for older VB6 and Win32 apps). Process is $150 per bot a month for unattended runs; Hosted Process is $215 with a Microsoft-hosted VM.
Generally available since May 2026. A model reads the screen of a Windows machine and works it with a virtual mouse and keyboard, at 5 Copilot Credits per step on a standard model or 15 on a premium one.
For any chain that starts and ends inside Microsoft 365, this is hard to beat.
The chain often leaves Microsoft for a property system, a PDF clause, or a bank portal.
Desktop flows and computer use run on Windows machines. Mac and Linux desks are not covered.
A release that renames a field breaks the flow until a maker repairs it.
Computer use bills every step, including retries.
In computer use, arithmetic belongs in a flow step, because the model should enter the number, not choose it.
Microsoft warns that a shared agent with maker-provided credentials lets anyone act with the author's access on that machine. Password fields in Citrix and other virtualized apps might not be supported.
LaunchAI works from the lease administrator's own desk on Mac, Windows, or Linux, with exact rules and per-action evidence. See Product →.
| Power Automate and Copilot Studio | LaunchAI | |
|---|---|---|
| Best at | Microsoft 365, Dataverse, and connected SaaS | Property and ERP screens, PDFs, portals, and exact math |
| Who builds | A maker comfortable with flows and selectors | The process owner, no prompts and no code |
| How it finds targets | Recorded selectors, or a model reading the screen | Layered screen reading; stops if unsure instead of guessing |
| Operating systems | Windows | Mac, Windows, Linux |
| Portals behind a named login | Stored credentials or the machine's session | The user's own signed-in browser |
| Exact math | Flow expressions and Excel | Deterministic rules, no model call |
| Human review | Approvals in cloud flows | A review step anywhere, clause and math on one screen |
| Evidence | Run history; session replay in preview | Before-and-after screenshots per action |
| Cost unit | Per user, per bot, or per credit step | License, no per-step meter |
CPI escalations in MRI: the index each lease names, the base month, floor, cap, and rounding, sometimes found only in the lease PDF.
The rent math must be exact, and capped increases should stop for review before anything is written.
The lease administrator works on a Mac.
Keep Microsoft for the Microsoft parts: Teams notices, SharePoint archives, Outlook distribution. LaunchAI writes each finished tenant notice to a watched folder, and a cloud flow files it and posts to Teams. Retire selector-based MRI steps only after a clean parallel cycle.
No. A desktop flow replays captured selectors. Computer use lets a model look at the screen each step, so it adapts, but bills per step and decides each action itself.
Not by design. Internal systems need a Cloud PC pool or your own registered Windows machine.
It can. LaunchAI reads the screen the same way for local or remote apps, but test your own delivery in a demo.
In LaunchAI, yes. Desktop flows and computer use need Windows.
No prompts. No babysitting. No Integrations.