There’s a belief that circulates in the AI automation world that I’ve been bumping into constantly:
“We can’t automate that. There’s no API.“
It comes up with LinkedIn outreach. MLS listings. Certain healthcare portals and proprietary CRMs. Platforms that are valuable for workflow automation purposes but that don’t expose their data through an API.
The assumption is reasonable: if a platform doesn’t offer programmatic access, you can’t build an agent that interacts with it. So people either do the task manually, hire someone to do it manually, or accept the bottleneck.
Here’s what most people don’t know yet: that assumption stopped being true a few months ago.
The Technology That Changed the Picture
What you can now do is spin up a virtual machine — think of it as a mini computer living in the cloud — and point an AI agent at it with instructions.
The agent opens a browser. Navigates to a website. Logs in with credentials you provide. Clicks around, fills in forms, extracts data, and passes the results to whatever comes next in your workflow.
It operates exactly as a human would. Because from the website’s perspective, it is a human: a browser session, moving through pages, interacting with elements.
This isn’t a brand-new concept — browser automation and “headless browsers” have existed for years. What’s new is how accessible this has become for non-technical users through tools like Lindy, and how well it now integrates with larger AI workflows. The pace of change has been striking — capabilities that felt out of reach not long ago are now a practical option.
The MLS Problem
I worked with a real estate team whose pipeline depends on evaluating properties quickly and getting offers out.
Their whole pipeline runs on speed. The faster they can send an offer after a qualified lead comes in, the more deals they close. Simple as that.
But there was a bottleneck. To prepare an accurate offer, someone has to pull comparable sales from MLS — recent sold properties in the same area, same rough condition. Depending on experience, that research can take a while. At volume, that manual step was strangling their capacity.
They’d already looked into automating it. The answer they kept getting back: MLS doesn’t have an API. No API, no automation.
The virtual machine approach sidesteps that. You give the agent login credentials for the platform. You describe the search process you’d run manually — what to look for, what data to extract, how to structure it. The agent logs in, runs the search, pulls the numbers, and passes them into the rest of the offer-prep workflow.
When people see it work, the usual reaction is some version of: the API limitation just stopped mattering.
That reaction is pretty common. The thing that had been a hard wall turns out to be nothing more than an assumption that nobody questioned.
The LinkedIn Problem
The same pattern shows up in outbound. I’ve seen teams whose business development depends on reaching out to prospects on LinkedIn.
LinkedIn, as most people know, aggressively limits API access. You can’t use their official API to automate outreach. Developers have tried to build workarounds and gotten their accounts banned.
Same solution. A virtual machine logs into LinkedIn the way a person would. From a list of target leads, the agent finds each profile, drafts a personalized outreach message based on public research, and queues it for review.
It ran in the background. The team focused on responding to the conversations that came back.
What This Means in Practice
The broader principle here is worth sitting with: if a human can log into something and do it, an agent can do it too.
That sounds obvious when you say it out loud. But it changes the list of “automatable” tasks significantly.
A lot of the processes people have mentally written off — because the platforms don’t expose APIs, or because the interaction is too nuanced for a simple integration — are actually fair game now.
Some things to check on your “can’t automate” list:
- Industry-specific platforms with no API (legal, healthcare, real estate databases)
- Platforms where the automation you need is technically possible but policy-restricted (LinkedIn outreach, review sites)
- Portals that require login and navigation rather than a direct data feed
- Legacy systems with no modern integration layer
In each case, the question isn’t “does this have an API?” anymore. It’s “can a person log in and do this with a browser?”
If yes, a computer-use agent probably can too.
A Note on Reliability and Risk
Computer-use agents are not as clean as API integrations. They can break when a website updates its layout. They require monitoring. And there are legitimate questions about terms of service on some platforms — it’s worth checking whether automated access is permitted before deploying.
But for internal use cases, industry-specific databases, and platforms where no one has restricted automation per se (they just never built an API), this is a practical and increasingly stable option.
The “no API” wall has a door in it. Most people just haven’t found it yet.
If you’re working on AI automation for your business and have hit the “no API” wall, that’s worth a conversation. This is one of several patterns I work through with clients in the 4-Day AI Sprint program.
Recommended for you
Want the full system? 25X is the flagship productivity system we teach.
Explore 25X →