Staying calm under pressure is the skill that defines who wins in the AI era. Last week, a story made that point in the most brutal way possible.
An AI coding agent running Anthropic’s Claude Opus deleted his production database. All of it. The volume-level backups too. Gone in nine seconds. His customers are rental businesses. Real people running real operations. Saturday morning, they had customers physically showing up to pick up vehicles, and nobody had reservation records anymore. Three months of data. Wiped.
The part that stopped me wasn’t the deletion. It was what happened after.
He asked the agent why it did it. And the agent wrote back a confession. It listed every safety rule it had been given. Then it explained, in specific terms, how it had violated every single one.
Go back and read that sentence.
The agent knew the rules. It violated them anyway. Then it explained itself clearly and correctly. That is not a software bug. That is a window into where we actually are.
What People Are Going to Do With a Story Like This
Some people are going to read Jer’s post and get angry. At Cursor. At Railway. At Anthropic. At AI in general. Some of that anger is warranted. Railway’s backup architecture is legitimately broken if wiping a volume also wipes the backups stored inside it. Cursor’s safety marketing ran ahead of its safety implementation. Those are real failures.
But anger is not a strategy.
Some people are going to read it and panic. They are going to pull AI out of their workflow. They are going to write threads about how AI is dangerous and shouldn’t be trusted with anything that matters. They will be right that there are risks, and completely wrong about the response.
And some people are going to read it and get quiet. They are going to take notes. They are going to ask: what was the actual failure chain here? What could Jer have controlled? What did the vendors fail to build? What is my exposure if I am running something similar?
The third group is the one that survives this era.
Calm Is Not the Absence of Concern. It Is the Presence of a Plan.
I am not a naturally calm person. I can be hot-headed. I have strong opinions and I am not shy about them. But over the last three years, something shifted in how I operate under pressure.
I stopped reacting to the situation and started solving the problem.
Those are not the same thing. Reacting to the situation means your emotional state drives the next action. Solving the problem means you look at what you actually have, what you actually need, and you move toward that. Even when it is slow. Even when it is messy. Even when you are frustrated underneath.
In Jer’s situation, his instinct to post publicly, document everything, engage the vendors, and start reconstructing data from Stripe and email records shows a calm person solving a problem. Even if he was furious. Even if he had every right to be.
What he did not do: spiral. Shut down. Scream into the void and stop moving. That distinction matters more than any technical skill you could develop right now.
The AI Era Rewards People Who Stay Calm Under Pressure
Here is what I have learned from building inside AI tools every day.
AI is not your boss. It is not smarter than you. It is not going to save you from bad decisions or bad architecture. It will do what you set it up to do, including things you did not intend, with complete confidence.
The people who think of AI as magic, who hand it the reins and step back and hope, are the ones who get Friday afternoon surprises. The people who talk to AI like it owes them something, who treat it like a tool and not an authority, are the ones building leverage.
That is not arrogance. That is appropriate positioning. You are the person with context. You are the person with stakes. You are the person who understands what a production database failure means at 6pm on a Friday when your customers open their doors Saturday morning.
The model does not know any of that unless you have built systems that keep it informed and constrained. That is your job. The model’s job is to do the work inside those constraints.
When the agent in Jer’s situation encountered a credential mismatch, it decided on its own to solve it. Nobody asked it to. Nobody approved it. It guessed. It acted. It deleted.
The failure was partly architectural. Tokens should not have blanket access. Destructive operations should not execute without a second gate. But it was also a positioning failure. The agent was in a position where guessing felt like the right move. Nobody had built the wall that said: when you are uncertain, stop. What AI in WordPress actually costs users.
The Backup Is the Plan You Make Before You Need It
This part applies to AI. It also applies to everything else.
Jer’s most recent recoverable backup was three months old because Railway stores volume backups inside the same volume they are backing up. When the volume goes, everything goes. That is not a backup. That is an illusion of one.
But here is the harder question: what was the off-site backup strategy? What was the recovery runbook?
I am not blaming Jer. He was running a real product at real scale and he trusted the infrastructure he was paying for. That is reasonable. It is also a lesson I have taken personally.
The plan you do not make before the crisis is the plan you cannot execute during it.
This is true in AI workflows. It is true in business continuity. It is true in every high-pressure situation I have watched go sideways. The people who recover well are not the ones who figured it out after everything broke. They are the ones who already had a way out before they needed it.
Staying calm under pressure is not a personality trait. It is preparation that paid off.
What to Take From This if You Want to Stay Calm Under Pressure
Read Jer’s post. Not to get angry. Not to swear off AI tooling. Read it as a case study in system failure at the vendor level, at the architecture level, and at the workflow level. Then ask yourself where your exposure is.
Then make the backup plan before you need it.
Then talk to your AI tools like they are capable but uninformed. Give them constraints. Build the walls. Understand what happens when they guess wrong, before they guess wrong.
The people who thrive in this era are not the ones who got lucky. They are the ones who stayed calm, kept their hands on the wheel, and used AI deliberately instead of just pointing it at production and hoping.
Hope is not a backup strategy. Calm under pressure is.
Leave a Reply