Why automating a broken process makes it worse

The most natural way to adopt automation is to take what you do today and have a machine do it instead. It feels safe. The process stays the same and only the executor changes, so there is nothing new to argue about and nobody has to change how they work. This is exactly why it is the most common approach, and exactly why it disappoints.

The problem is not the automation. The problem is the assumption hiding underneath it: that the process, as it runs today, is worth preserving. In an established company it usually is not. Not because anyone was careless, but because processes in old companies are not designed. They accumulate.

Processes are fossils

Look closely at any process that has run for fifteen years and you will find the sediment of every constraint the company has ever lived under. There is an approval step because someone once sent the wrong price to an important customer, in an email client that no longer exists. There is a weekly batch because the system it feeds used to fall over if you loaded it during the day. The same information is entered twice because two departments once had two systems, and although the systems were merged years ago, the habit was not.

Every one of these steps made sense when it was added. Nobody went back to ask whether it still does, because processes do not have owners the way products do. They have operators, and an operator's job is to run the process, not to question it.

Automate a process like this as it stands and you do not just inherit the fossils. You set them in concrete. A step that used to be a habit becomes a requirement, written into software, with a price attached to ever changing it. The project that was supposed to modernise the company has quietly made its past mandatory.

Speed makes everything more of what it is

Automation is an amplifier. It makes whatever you point it at happen faster, more often, and at a lower cost per instance. It has no opinion about whether the thing it amplifies is good.

If a process occasionally produces the wrong result, the automated version produces the wrong result at scale, and each error travels further before anyone catches it. This is worse than it sounds, because of a quality control mechanism that never appears in any process description: the bored human. A person doing repetitive work is also, quietly, an inspector. They notice the order that looks strange, the amount with one zero too many, the name that does not match the account. Nobody wrote that inspection down, so nobody knows it is load bearing. You find out it existed when it is gone.

Delete before you accelerate

The right question to ask about a process is not what it is, but what it is for. These sound similar and are not. What the order process is: eleven steps, four systems, three approvals. What it is for: the customer gets what they ordered, the company gets paid, and someone finds out early if either is at risk. Everything in the first description that does not serve the second is a candidate for deletion, and there is always something.

So the sequence matters: first shorten, then speed up. Take the steps one at a time and ask what breaks if this disappears. Some steps defend themselves immediately. Some turn out to protect against a risk that stopped existing years ago. And for some, the honest answer is that nobody knows, which is not a reason to keep the step. It is a reason to find out, because a step nobody understands is not process. It is superstition.

The best automation projects are, in large part, deletion projects. The fastest step is the one that no longer exists. It also never breaks, costs nothing to maintain, and never has to be explained to a new employee.

What agents change

There is a second reason to redesign now rather than automate as is, and it is specific to this moment. The old generation of automation could only handle structure. Fixed forms, rigid categories, one happy path. So for thirty years companies bent their processes to fit the machines. Make the customer fill in the form. Force every request into one of six types. Push everything unusual into a queue for a human. A surprising amount of what is ugly about your current process is scar tissue from making it automatable with the tools of 2009.

Agents remove that constraint. They can read the email the way the customer actually wrote it, handle the case that fits no category, and ask a question when something is unclear. The distortions you added to survive the old machines are no longer load bearing. A process redesigned today can be shaped around the work itself rather than around the limitations of whatever software processes it.

But the same flexibility cuts the other way. Because agents can cope with a messy process, it has never been easier to automate the mess exactly as it is. The technology will no longer force you to think first. The thinking was most of the value.

Knowing where the bodies are buried

Here established companies hold a real advantage, and it is worth naming, because they are so often told the opposite. Redesigning a process safely requires knowing why each step exists. In an established company, someone does. Every strange rule has a story, and the person who remembers the story usually still works down the corridor. A startup redesigning its process is guessing. A twenty year old company doing the same is remembering, which is faster, safer, and far more likely to keep the steps that were protecting something real.

That window does not stay open forever. The people who remember are not staying forever, which is a problem large enough to deserve its own essay. But while they are there, the work is not complicated. Ask what the process is for. Delete what does not serve it. Then, and only then, make it fast.