“But It Doesn’t Have an API” — The Objection That No Longer Applies

Every time I run a workshop on AI automation, the same objection comes up.

We get deep into what’s possible — email automation, meeting prep agents, content workflows — and then someone in the group points to the thing they actually care about.

“Can AI handle LinkedIn outreach? Because I’ve heard the API doesn’t allow it.”

Or: “Our booking system is old. Nothing integrates with it.”

Or: “We pull data from the MLS, but it’s all behind a login. There’s no feed, no API, nothing.”

These are real objections. The tools people want to automate often don’t cooperate. They were built before modern API standards, or they deliberately restrict access, or they’re just a website that expects a human to log in and click around.

For a long time, that was the end of the conversation. If there’s no API, there’s no automation.

That changed.

What Computer Use Actually Does

Most automation tools work by connecting to an application’s API — a structured way for software to communicate with other software. When LinkedIn exposes an API, Zapier can use it. When your CRM has an API, Lindy can read and write to it.

But a lot of software doesn’t have a usable API. And a lot of the ones that do have one restrict what you can actually do through it.

Computer use agents work differently. Instead of talking to the software beneath the interface, they interact with the interface itself. They see the screen like you see it. They click buttons. They read what’s displayed. They type into fields. They scroll, navigate, wait for pages to load.

The frame that makes this click: if a website makes a human log in, an AI agent can too.

It’s not magic. It’s the same set of actions you’d take — just automated, consistent, and running whenever you need it to.

The Real Estate Client

Last fall, I was working with someone who needed to track MLS real estate listings daily. New properties, price changes, listings that matched specific criteria — the kind of data that drives their lead flow.

The problem: no API. No data feed. The MLS is behind a login, and access is restricted. The only way in was to log in and click around manually.

A contact of mine who builds similar systems described it well: they built a virtual machine that mimics a human logging in, clicking through the interface, and extracting the data. Properties matching the search criteria get dumped into a Google Sheet. That triggers the next layer of automation — lead scoring, follow-up workflows, alerts.

It runs 24/7. Lead flow went up dramatically. No human needs to remember to check the MLS every morning.

The same technology that felt experimental a couple years ago is now something you can set up in an afternoon.

What’s Now Automatable

When I talk about computer use at workshops, the reaction is usually the same: people immediately start mentally applying it to the thing they’d given up on.

Here’s a sample of what I’ve actually built or seen built with computer use agents:

LinkedIn outreach. The official LinkedIn API restricts a lot of what people want to do — sending connection requests, messaging, profile scraping for research. Computer use gets around it. The agent logs into LinkedIn and operates it like a person would, just at scale and without manual effort.

MLS and real estate data. Login, filter, extract, deposit into wherever you need it.

Restaurant reservations. Log in to OpenTable, Resy, or whatever the restaurant uses. Find a table at the right time. Book it. No API needed.

Barber and salon appointments. The booking system doesn’t need to integrate with anything — the agent just uses it the way any customer would.

Grocery ordering. I’ve set this up with HEB — the agent logs in, adds items to the cart based on a list I’ve specified, proceeds to checkout, and sends a confirmation email.

Every one of these used to be in the “can’t automate” bucket. Not because automating them is technically complex, but because people assumed you needed an API and these sites don’t have one.

The Shift in Thinking

There’s a mental model worth updating here.

The old model: the question for whether something can be automated is “does this tool have an API?” If yes, maybe. If no, no.

The new model: the question is “can a human use this website?” If yes, an agent can use it too.

This sounds like a small change, but it’s large in practice. A huge portion of the web — and a huge portion of the specific workflows that create friction for people — runs on websites that a human can navigate but that no integration tool can reach. That category is now automatable.

The caveat worth naming: computer use is slower than API access. The agent is actually loading pages and waiting for elements to appear. For tasks you do once a week, that’s fine. For tasks requiring thousands of operations in rapid succession, API access is still faster when you can get it.

But for the tasks that matter — the daily login to pull data, the weekly booking task, the outreach workflow that’s been stuck in “can’t automate this” — computer use gets them done.

Getting Started

If you have a workflow that’s been stuck on the “no API” problem, here’s how to think about approaching it:

Map the human steps. What does a person actually do when they complete this task? Log in to what URL. Click what button. Fill in what field. Search for what. Copy what to where. The agent will follow the same sequence.

Decide on the trigger. What should kick off this workflow? A scheduled time? A new row in a spreadsheet? An incoming email? The computer use action is usually the middle of a chain, not the whole thing.

Decide where the output goes. The agent extracts or acts on data — where does it go after? A Google Sheet, an email, a Slack message, a CRM record?

Build it small first. A single login and a single action is enough to prove the pattern works. Don’t try to automate the whole workflow on the first pass.

The constraint most people have been living with isn’t actually a constraint anymore. The question was never whether the site has an API. It’s whether a human can use it.

P.S. Computer use agents are one of the more advanced setups we walk through in the Two Hour Workday program.