For years, the answer to "can we automate this?" started with "is there an API?" If the answer was no — which it usually was for internal admin panels, legacy ERPs, and that one app the ops team has used for a decade — the conversation ended there.
That's no longer the full picture.
What changed
Computer-use agents — AI that operates software the way a person does, by looking at the screen and clicking and typing — have gotten good enough to navigate real applications with no documentation and no API shortcut. Give one a goal and a login, and it will find the right menu, fill in the right form, and complete a multi-step task on software it has never seen before.
We've tested this directly on our own internal tools: an agent given nothing but a plain-language goal completed it correctly on the first attempt, no walkthrough required.
Why this matters for internal tools specifically
Internal tools are exactly the category that used to be hardest to automate — no public API, no vendor incentive to build one, and rarely enough budget to justify custom integration work. Computer-use agents change the cost-benefit math for this whole category. Legacy ERPs, vendor portals without an API, that one internal app nobody wants to touch — all of it just became more reachable, without months of integration work.
The catch
It's not magic, and it's not a replacement for a proper API when one exists or is worth building. Long sessions, ambiguous state, and anything requiring real judgment across many steps are still failure modes worth respecting. This is a new option in the toolkit, not a universal one.
What we're doing with it
For clients with software that was previously "too old" or "too custom" to automate affordably, this reopens the conversation. The question is no longer only "is there an API?" — it's "can an agent operate this well enough to do the job?" For a growing number of workflows, the honest answer is yes.