Hired right before a botched system rollout. Quit when…
0 reactions
Hired right before a botched system rollout. Quit when they answered our concerns with “just add AI on top.”
*(Yes, I used AI to help me write this so it reads better — I’m a drawings / numbers / people person, not much of a writer. But the context is all real and all mine.)*
Got hired earlier this year, right before my company rolled out a big new ERP system. Figured I was joining something solid.
Nope.
The whole implementation was a mess. The people who were supposed to know the system half the time didn’t know what was going on. Processes that were supposed to run automatically just… didn’t. I found out I’d been trained to pull information from the wrong place, so I’d been giving customers outdated numbers without knowing it.
Then they asked me to test a new customer-facing part of the system. I found a pile of bugs in about an hour. They launched anyway. No follow-up. It felt political — push it live so nobody upstairs has to admit it wasn’t ready. Meanwhile there was a whole pipeline of other stuff being built by people who didn’t really seem to know what they were doing.
The result: so much cleanup got created that I was spending maybe 30% of my time on my actual job and 70% firefighting other people’s mess.
So my team and I took it all the way up — straight to the CEO and HR, with our department head backing us. We said plainly: a ton of work has been created that nobody wants to own, so it just gets pushed around between us. The answer was short — basically “that’s your responsibility” — but hey, they were *concerned*, and they *promised to work on it.*
A couple weeks later, the fix arrives. An email lands: they’re rolling out an AI that will auto-generate customer orders for us. And the best part? It’s going to *make our day easier and more efficient.*
Here’s the thing. The AI reads incoming emails, matches item numbers and customer details, and auto-creates the order in the system. You just “approve” it. Sounds great. Except the whole thing rests on our data being clean — and our data is a graveyard. Tons of discontinued items, wrong prices, entries that don’t match reality, all built by different people with no shared structure or plan. That’s exactly the concern we’d raised. Garbage in, garbage out — except now an AI confidently generates the garbage and a human “approves” it under time pressure.
So you can see how this goes: the AI auto-creates an order against a phantom item or an old price, it lands in the approval queue looking legit, someone rubber-stamps it because they’ve got fifty of them, and now there’s a wrong order in the system. And whose job is it to catch and clean that up? Mine. The same 70% pile, except now it’s faster to generate and harder to trace, because a layer of “automation” sits between the mistake and the human.
And that’s just one of them. The deeper problem is what the job actually becomes. Before, you’d read an email and type the order in — one task, and while you typed you understood what you were entering. Now you sit and *validate* what a bot typed: re-reading every field to check whether the right information actually landed in the system.
That’s not less work. Reviewing someone else’s input for errors is often slower and more mind-numbing than just doing it yourself, because you can’t trust any of it and you can’t see how it got there.
And nobody’s been clear on where it stops. We specialize in assembling parts from different manufacturers into one solution for the customer — so does the AI just create the customer order, or does it auto-generate purchase orders to the manufacturers too? Because if it does, a wrong item number doesn’t just sit in a quote, it goes out the door and we’ve actually ordered the wrong thing.
Nobody answered that. I’m not sure anyone asked.
It gets worse when you know our customers. They regularly send us their old orders to get a fresh quote on parts they bought before — sometimes orders that are *20 years old*. So the AI is now reading emails full of decades-old item numbers, matching them against a database full of discontinued products and wrong prices, and confidently generating orders from that. The one situation that most needs a human who knows the catalog and the history is the exact situation they want to hand to a bot.
They had the intention right — less typing, faster response, more time for customers. But nobody actually listened to the two things that matter: the data foundation is unreliable, and nobody owns the mess. You can’t automate your way out of a structure problem. Automation on top of bad data doesn’t reduce the work, it just relocates it and adds a layer of plausible deniability.
This wasn’t a message that got lost on its way up. The top of the company heard it directly, looked at it, and answered with a shrug and an AI rollout.
So I quit. No new job lined up. Because I’m not going back to spend my career as the human backstop for an automation pipeline built on a foundation nobody will fix — cleaning up errors that are now generated at machine speed, while being told it’s my responsibility and that things are getting “more efficient.”
Sometimes the move is to leave before the thing you warned them about becomes your full-time job.
source
*(Yes, I used AI to help me write this so it reads better — I’m a drawings / numbers / people person, not much of a writer. But the context is all real and all mine.)*
Got hired earlier this year, right before my company rolled out a big new ERP system. Figured I was joining something solid.
Nope.
The whole implementation was a mess. The people who were supposed to know the system half the time didn’t know what was going on. Processes that were supposed to run automatically just… didn’t. I found out I’d been trained to pull information from the wrong place, so I’d been giving customers outdated numbers without knowing it.
Then they asked me to test a new customer-facing part of the system. I found a pile of bugs in about an hour. They launched anyway. No follow-up. It felt political — push it live so nobody upstairs has to admit it wasn’t ready. Meanwhile there was a whole pipeline of other stuff being built by people who didn’t really seem to know what they were doing.
The result: so much cleanup got created that I was spending maybe 30% of my time on my actual job and 70% firefighting other people’s mess.
So my team and I took it all the way up — straight to the CEO and HR, with our department head backing us. We said plainly: a ton of work has been created that nobody wants to own, so it just gets pushed around between us. The answer was short — basically “that’s your responsibility” — but hey, they were *concerned*, and they *promised to work on it.*
A couple weeks later, the fix arrives. An email lands: they’re rolling out an AI that will auto-generate customer orders for us. And the best part? It’s going to *make our day easier and more efficient.*
Here’s the thing. The AI reads incoming emails, matches item numbers and customer details, and auto-creates the order in the system. You just “approve” it. Sounds great. Except the whole thing rests on our data being clean — and our data is a graveyard. Tons of discontinued items, wrong prices, entries that don’t match reality, all built by different people with no shared structure or plan. That’s exactly the concern we’d raised. Garbage in, garbage out — except now an AI confidently generates the garbage and a human “approves” it under time pressure.
So you can see how this goes: the AI auto-creates an order against a phantom item or an old price, it lands in the approval queue looking legit, someone rubber-stamps it because they’ve got fifty of them, and now there’s a wrong order in the system. And whose job is it to catch and clean that up? Mine. The same 70% pile, except now it’s faster to generate and harder to trace, because a layer of “automation” sits between the mistake and the human.
And that’s just one of them. The deeper problem is what the job actually becomes. Before, you’d read an email and type the order in — one task, and while you typed you understood what you were entering. Now you sit and *validate* what a bot typed: re-reading every field to check whether the right information actually landed in the system.
That’s not less work. Reviewing someone else’s input for errors is often slower and more mind-numbing than just doing it yourself, because you can’t trust any of it and you can’t see how it got there.
And nobody’s been clear on where it stops. We specialize in assembling parts from different manufacturers into one solution for the customer — so does the AI just create the customer order, or does it auto-generate purchase orders to the manufacturers too? Because if it does, a wrong item number doesn’t just sit in a quote, it goes out the door and we’ve actually ordered the wrong thing.
Nobody answered that. I’m not sure anyone asked.
It gets worse when you know our customers. They regularly send us their old orders to get a fresh quote on parts they bought before — sometimes orders that are *20 years old*. So the AI is now reading emails full of decades-old item numbers, matching them against a database full of discontinued products and wrong prices, and confidently generating orders from that. The one situation that most needs a human who knows the catalog and the history is the exact situation they want to hand to a bot.
They had the intention right — less typing, faster response, more time for customers. But nobody actually listened to the two things that matter: the data foundation is unreliable, and nobody owns the mess. You can’t automate your way out of a structure problem. Automation on top of bad data doesn’t reduce the work, it just relocates it and adds a layer of plausible deniability.
This wasn’t a message that got lost on its way up. The top of the company heard it directly, looked at it, and answered with a shrug and an AI rollout.
So I quit. No new job lined up. Because I’m not going back to spend my career as the human backstop for an automation pipeline built on a foundation nobody will fix — cleaning up errors that are now generated at machine speed, while being told it’s my responsibility and that things are getting “more efficient.”
Sometimes the move is to leave before the thing you warned them about becomes your full-time job.
source
| 4,000 mlx | |
| 8 | |
| 0 |