Should we replace what we have?

Only where a proper way in now exists, which is worth checking before assuming. Many older systems have since gained one, and where that is true the replacement is more reliable and easier to maintain. Where it is not, removing the screen-driving means replacing the system underneath, which is a much larger conversation.

Why does it break so often?

Because it depends on a screen staying arranged as it was when the path was recorded. A moved button, a renamed field, an added confirmation or a slow morning all interrupt a sequence while the task itself is unchanged. The fragility is inherent to operating an interface rather than a sign of poor implementation.

Is the newer AI version more reliable?

More adaptable, which is not the same as more predictable. A system choosing where to click can cope with a moved button that would have stopped a recorded path, and it can also click something unintended in a way a fixed path never would. You trade a brittle failure for a wider range of behaviour.

Glossary

Robotic process automation (RPA)

RPA drives software the way a person does, clicking and typing through screens, which is what makes it work with systems that offer no other way in and what makes it fragile.

In plain terms

Software that uses other software the way a person would: moving to a field, clicking, typing, pressing enter. It exists because some systems offer no other way in, particularly the old ones a business depends on. It works, and it breaks whenever anything on the screen moves, which is the whole story in one sentence.

Why it matters

Because a great deal of it is still running, quietly, in large organisations, and any conversation about replacing it has to start from why it was chosen. It was rarely the preferred approach. It was what remained once a system turned out to have no other means of access, and that constraint has not gone away because the tooling around it improved.

How it works

It operates the interface rather than going underneath it, which is both the point and the flaw. Where a system exposes a proper way in, going through that is faster, more reliable and easier to reason about; where it does not, the screen is the only surface available and clicking through it is the only option that exists. The choice is usually made for you.

Fragility follows directly from that, and it is not a quality problem. A button moving, a field being renamed, an extra confirmation appearing after an update, a slower load on a busy morning: each of these breaks a sequence that was recorded against a particular arrangement of a screen. The underlying task did not change, and the automation still stops.

Its home is systems that predate the expectation of being connected. Older finance, records and administrative platforms frequently offer no other route, and replacing them is a multi-year proposition, so driving their screens is a rational compromise rather than a mistake. Understanding that is the difference between removing it and being surprised by what it was holding together.

The newest AI tools have arrived at the same compromise from the other end, which is the part worth noticing. Letting a system drive a screen because a piece of software offers no other way in is exactly the old reasoning, now with a model choosing where to click rather than a recorded path. It is more adaptable to a moved button and it shares the same underlying dependence on the interface staying put.

Two ways into the same system

The reason this matters for anybody inheriting an estate is that the left-hand column tends to be read as a legacy of poor decisions, and it usually is not. Somebody needed a figure to move from one system to another every morning, the receiving system had a proper way in, the sending one did not, and the choice was between driving its screen and waiting for a replacement programme that was already years out. Driving the screen was correct. What went wrong afterwards is a different thing: the arrangement was never revisited when the old system finally gained a proper route in, because nothing prompts that review and the automation kept working well enough. So the useful first question about any of this is not whether to remove it but whether the constraint that caused it still exists. Frequently it does not, and the replacement is a smaller job than the original.

Seen in the wild

A browser-resident tool that scrapes structured data off pages and moves information between web applications, which is the same shape at a smaller scale.
Bardeen
A platform whose durable advantage is the long tail of applications nothing else speaks to, which is the alternative when a proper way in does exist.
Zapier
A self-hostable platform whose visual steps accept real code, used where an older system needs coaxing rather than clicking.
n8n

Common misconceptions

People assume

AI has made it obsolete.

In fact

AI has arrived at the same problem by the same route. Letting a system drive a screen because there is no other way in is precisely the old reasoning, and the newer tools are better at coping with a moved button rather than freed from depending on the screen. The constraint was never the technology.

People assume

It was a poor engineering choice.

In fact

It was usually the only choice on offer. The systems it drives tend to predate any expectation of being connected, and replacing them is a programme rather than a task, so clicking through their screens was a rational response to a constraint somebody else created decades earlier.

Telling them apart

Robotic process automation vs Computer use

Robotic process automation

A recorded path through a screen, followed exactly.

Computer use

A system deciding where to click as it goes.

Same surface, different steering. One breaks when the button moves and the other tries to find it, which is an improvement in robustness rather than a change of approach.

Questions

Should we replace what we have?: Only where a proper way in now exists, which is worth checking before assuming. Many older systems have since gained one, and where that is true the replacement is more reliable and easier to maintain. Where it is not, removing the screen-driving means replacing the system underneath, which is a much larger conversation.
Why does it break so often?: Because it depends on a screen staying arranged as it was when the path was recorded. A moved button, a renamed field, an added confirmation or a slow morning all interrupt a sequence while the task itself is unchanged. The fragility is inherent to operating an interface rather than a sign of poor implementation.
Is the newer AI version more reliable?: More adaptable, which is not the same as more predictable. A system choosing where to click can cope with a moved button that would have stopped a recorded path, and it can also click something unintended in a way a fixed path never would. You trade a brittle failure for a wider range of behaviour.