Imagine you wake up tomorrow and AI is gone. Not throttled. Not regulated. Gone. Every model offline. Every assistant dark. Every pipeline that quietly depended on a foundation model somewhere in the stack, broken. The world's AI dependency, exposed in a single morning.
For about a week, half the workforce would freeze. Not because they're incompetent. Because they have built their daily output on a layer they no longer notice using. Marketers staring at a blank page they used to fill with a prompt. Analysts unable to summarize the reports they used to feed into a chat window. Customer service queues exploding because the bots that handled tier one are dead. Engineers watching their pull request velocity drop by half because their copilot is gone. Recruiters with no first-pass screen. Designers without backgrounds-removed in one click. Lawyers with thirty more hours of reading per case.
The first week would look like a soft civilizational stutter. Not catastrophic. Just slower. Heavier. More expensive. Like a country running on the fuel reserves it forgot it had stopped maintaining.
The depth of AI dependency only shows when you remove it.
Once the panic subsides, something more interesting begins to surface. Companies start asking a question they hadn't asked clearly in three years: what was that AI actually for?
Most of the answers, when honestly given, are uncomfortable.
A surprising amount of corporate AI use isn't generating value. It's generating volume. More emails written, none of them better. More content shipped, none of it more memorable. More reports summarized, more dashboards refreshed, more meetings transcribed nobody reads. AI made it cheaper to produce work, so we produced more work, and called it productivity.
An even larger share of AI use is solving problems that other AI use created. AI writes too many emails, so AI summarizes them. AI generates more content than humans can review, so AI ranks it. AI creates so much noise that other AI is hired to filter the noise. Take all of that away and a strange thing happens. The noise drops. The filtering need drops with it. The system gets quieter and most people don't actually miss it.
An AI disappearance wouldn't return us to 2022. It would force a forced audit of what we'd been pretending the technology was for.
Three things would survive the audit.
The genuinely useful AI uses would emerge from the wreckage looking obvious. The medical imaging system catching tumors radiologists miss. The customer service bot that handles 80% of tier-one questions in 12 seconds instead of 12 minutes, freeing humans for the hard cases. The translation layer that lets a Polish factory worker file a safety complaint in real time. The accessibility tools that let a blind student read a PDF without help. These survive any reasonable audit because the value is unambiguous.
So would the deep skill-amplification uses. The lawyer using AI to read fifty documents in the time it used to take to read five, and then doing better legal work because of it. The strategist running ten scenarios in an afternoon instead of three over a week. The product designer testing twenty layouts instead of two. These aren't AI replacing human judgement. They're AI extending human judgement. Different category.
And so would the boring back-office plumbing. The fraud detection. The supply chain optimization. The maintenance scheduling. The unsexy systems that have been quietly running on machine learning for fifteen years and that nobody calls AI because the word AI got hijacked by chatbots in 2023.
What dies is the AI dependency on busywork.
Most of what would die in week two is the inflation layer. The work created by AI to be consumed by other AI. The performative productivity. The volume-without-value content treadmill. The roles that exist because AI generated more output than humans could process and now we need humans (or more AI) to process it.
None of that comes back. And most companies wouldn't actually want it back, once they saw what their organization looked like without it.
The people who really get hurt aren't the ones losing the inflation layer. They're the ones who have spent two years offshoring their thinking to a system. Junior strategists who never built the muscle of forming an opinion before asking a model for one. Marketers who can't write a paragraph cold. Engineers who can scaffold but can't reason. The skill atrophy is real, and it's the part of the AI dependency that wouldn't heal in a week or a quarter or maybe a year.
The AI dependency question is one you can ask now.
The thought experiment isn't really about a disappearance. It's about the difference between two kinds of dependence. The kind where a tool extends what you can do. And the kind where a tool replaces what you used to be able to do.
The first kind survives a disappearance. The second kind doesn't.
The good news is you don't need an actual disappearance to figure out which kind your organization is running on. You just need to ask the question now. Out loud. With the people who would have to do the work without the tool.
What you find out will be more useful than any AI strategy document.
Want to bring these ideas to your organisation?
Ferry speaks about AI, Brand Humanizing, and the future of work at conferences and corporate events worldwide. Book him for your next event.