We run performance review cycles at Remote in 48 hours using AI. What used to take eight weeks, running the cycle for almost 2,000 people, now takes two days.
People ask how we prepared for that. Honest answer: We didn’t prepare for AI at all. We had monthly check-ins, consistent documentation, and calibration sessions. We did this because that’s just what running a company well looks like, not because we were building toward some AI future. When the technology arrived, we were able to compress the whole cycle overnight.
Here’s what most people need to learn: When AI “fails” at a company, it’s because the AI found a problem already present.
A company automates revenue projections on top of unvalidated sales data and gets confident-looking nonsense. A payroll team deploys AI for compliance on top of misconfigured rules and it breaks on edge cases.
In both cases, the problem existed long before the AI did. Humans were working around it, filling gaps by hand, absorbing the errors. AI just made it impossible to hide.
Air Canada learned this in public. Their chatbot gave a passenger wrong information about bereavement fares and a court held them liable. Many read it as an AI cautionary tale. I read it differently: Their policy information was inconsistent before the chatbot existed. The AI surfaced that in weeks instead of letting it rot for years.
The conventional wisdom about AI is to slow down, build the foundation first, deploy later.
I think that’s wrong. If your process is broken, waiting doesn’t fix it. Deploying AI does, because now the brokenness is visible, measurable, and impossible to ignore. Every AI failure is a free audit of your operation. The companies moving fastest are discovering risk that was already there, while their competitors let it compound in the dark.
A June 2026 IBM study found that 70% of organizations have teams deploying AI tools more quickly than leadership can track it. That gets framed as a crisis, but it’s what adoption looks like. The fix is for leadership to speed up.
Yes, stakes matter. A performance review affects someone’s career. Payroll affects whether they can pay rent. In those domains, when you use AI, keep a human involved in the output while you build confidence. That’s how we did reviews. But “keep a human in the loop” and “don’t deploy yet” are very different things. One is engineering. The other is fear.
So before you deploy AI to a process, ask yourself: Do I trust it? If yes, go. The speed compounds. If no, go anyway. You’re about to find out what’s broken, years earlier than you would otherwise.
The only real failure mode is standing still while everyone else learns at 10 times your speed.
*Job van der Voort is *CEO and cofounder of Remote.