🛠️ Why Replacing Humans with AI Will Fail
Lessons from the field, the call center, and the chaos of undocumented software
The Plan to Replace Human Workers With AI Will Fail.
Bold claim? Maybe. Said with my whole chest? Absolutely.
We’re living in a time where executives and marketing teams genuinely believe AI can replace entire departments of human workers. And I get it — the idea is seductive. “Just feed the AI all the documentation, and boom! No more tech support.”
But here’s the problem with that dream: It completely misunderstands what real expertise looks like in the field.
Back before chatbots and generative tools were trendy, if you needed to fix something — really fix something — you had to talk to a human expert. Someone like me.
I worked for an carrier-grade tech vendor. Most of our technical documents? Locked behind a paywall. They still are. Even today, GPT can’t access or synthesize the kind of engineering documentation that was part of my everyday life. In many ways, we were the LLMs — the human Language Application Layer.
We were trained to ask:
- “What version are you on?”
- “Have you recently updated?”
- “When did this issue first appear?”
- “Do you have a backup?”
Not because we liked repeating ourselves, but because context changes everything. What works in 2.1 might break entirely in 3.0. And if a button disappears or a UI changes? That’s not a minor issue — that’s DEFCON 2 for the support team.
A customer once tried to get me fired… Because I said “You appear to be at capacity.”
He heard it as “You're unsupported.”
I handled it. De-escalated. Pinged my manager while troubleshooting. Just like clockwork, 15 minutes later:
“Did you tell the customer their network was over capacity?”
“No. I said ‘appears’ — and then confirmed it was.”
We played the call. I had the receipts. I was cleared.
Millennials didn’t invent the habit. We just perfected it.
Which brings me back to AI.
When I ask an AI assistant for help, I can tell it’s trying. Really. But what it can’t do — what it will fail to do — is this:
- It doesn’t ask for your version.
- It doesn’t know if the UI changed last week.
- It doesn’t know if your team updated the docs.
And most of the time? The docs aren't updated.
Just look at the image above. That’s the official Onshape FeatureScript documentation (left) versus the actual UI (right). Doesn’t match. And if AI’s pulling from the docs? Then AI’s already wrong.
The machine can only be as good as the dataset. And the dataset is often outdated garbage.
Meanwhile, the humans you’re trying to replace? We weren’t just reading docs. We were:
- Debugging live production systems
- Translating abstraction into action
- Absorbing panic and guiding people through fire
- Tracking subtle version-to-version quirks that no one documented
You’re not just replacing a job. You’re removing the only layer that adapts to change.
Until your AI knows to ask “what version are you on?” — don’t talk to me about replacing humans.