Most AI integrations work through APIs. You connect a service, Companion calls the API, the task is done. Clean, fast, reliable.
But some platforms don't have APIs for everything you'd want to automate. LinkedIn, for example, doesn't expose a public API for posting content. So how does Companion post to LinkedIn?
It uses Computer Use — the AI controls the browser like a person would.
What happened
We asked Companion to post on LinkedIn using an image from a specific folder, with a caption about the image being a test post.
No LinkedIn MCP. No API key. Just a browser and a prompt.
Here's what Companion did:
- Opened its own browser window — separate from any browser you have open
- Navigated to LinkedIn automatically
- Retrieved the image from the specified folder
- Composed the post with the caption we described
- Published it
The whole sequence happened autonomously. We watched the cursor move, the browser navigate, the image upload, the text appear, the post button click.
The post appeared on LinkedIn.
"Like my computer is haunted by the spirit of Claude"
That's a direct quote from the person running the demo. It's the most accurate description of what Computer Use looks like when you first see it.
The browser opens on its own. The cursor moves. Forms fill in. Buttons click. There's no human behind the keyboard — the AI is doing it.
It's not a screen recording. It's not a macro. It's the AI interpreting what it sees on screen and taking the next action, step by step, until the task is complete.
Why Computer Use matters
APIs cover the well-supported automation cases. Computer Use covers everything else.
Platforms with no API: LinkedIn posting, some government portals, legacy enterprise software, internal tools built on old frameworks
Actions APIs don't expose: interacting with visual elements, navigating dynamic UIs, handling multi-step wizard flows, filling CAPTCHA-adjacent forms that require human-like behavior
One-off tasks: you need to do something once in an app you don't use regularly enough to build a proper integration for. Computer Use handles it without setup.
The result: Companion can automate almost anything with a UI, not just the services that publish developer APIs.
The tradeoffs
Computer Use is slower than API calls. An API call takes milliseconds. A browser-based task takes seconds or minutes, because Companion has to navigate the UI at human-readable speed.
It's also less reliable than APIs for tasks that require precise data handling — reading structured data from a UI is harder than reading a JSON response.
When to use Computer Use: tasks that have no API, one-off automations, or workflows in apps that change their UI infrequently.
When to use MCP integrations: tasks you run regularly, anything requiring precise data, high-volume automations.
Most power users combine both. Use MCP for Gmail, Calendar, Notion, Slack — the apps they work in daily. Use Computer Use for the edge cases that don't fit.
What else Computer Use can do
We've shown it installing software, posting to social media, and filling web forms. Other examples from real Companion sessions:
- Navigating a booking system to reserve a table at a restaurant
- Filling out a government form with personal details
- Interacting with a desktop app that has no API (resizing images in Photoshop, entering data into accounting software)
- Scraping information from a website that blocks API access
The mental model: if a person with a browser could do it, Companion with Computer Use probably can too.
Download Companion and try a Computer Use task. A good starting point: "Open YouTube and search for [topic]. Tell me the titles and view counts of the first 5 results." Watch it work.