Just over a year ago, the leading AI experts said: "Learn AI and your job is safe." I felt like the lone skeptic saying: Well, that's like saying "Learn to swim and the Titanic won't sink." It's why I started this newsletter - to shed some light on the monster under the bed.
Now? Plot twist. The same experts have flipped 180 degrees. They're looking at the data rolling in and saying "Actually, yeah, jobs are definitely going away." The very people who told everyone not to worry are now... worrying. Not because new jobs aren't going to be created, they will be at some point, but it’s going to take too long to replace those being replaced now to keep the economy running smoothly.
Data from emerging studies reflect their new doom:
Meanwhile, I'm over here having my own plot twist. After watching AI fail spectacularly at simple tasks it was supposed to master by now, I've gone from "AI doom" to "humans-are-more-essential-than-ever" optimist. It's like we're all playing musical chairs with our predictions. While the experts are now panicking about job displacement, I'm seeing something they're missing: AI can't actually deliver on its promises and today's promises are just as good as Elon's level 5 self-driving car prediction in 2017, stating it will be here in 2019, AI has a LOOOOONG way to go before it’s going to replace roles en masse and reduce the overall workforce size permanently.
After 10 years building and promoting AI projects, seeing how the landscape is evolving, and two years diving deep into these generative tools, I've discovered the ultimate irony: The technology everyone feared would replace us is desperately dependent on us. The experts went from "AI will empower workers" to "AI will eliminate workers". The reality is "AI can barely function without workers." This batman joke clearly shows it. Batman has a bat mobile, bat plane, bat phone and a batarang... and all the latest AI models fail understanding the joke (at the time I wrote this). It’s easy for humans to decode… but let’s look at how ChatGPT 4.5 does with it as of June 30th 2025:
Minimize image
Edit image
Delete image
A fresh Carnegie Mellon study shows AI agents fail over 70% of the time in complex workflows. Gartner is now predicting more than 40 percent of agentic AI projects are going to be cancelled before 2027 ends as a result of costs, unclear business value, risks. .Even OpenAI's own system cards (pdf) admit their models hallucinate, sandbag, and occasionally "sabotage" tasks under pressure.
Here's the delicious irony: While experts flip-flop between "AI will save jobs" and "AI will destroy jobs," the actual AI is out there failing spectacularly at both. Companies like Klarna - who made headlines replacing 700 jobs with AI are quietly rehiring humans because, surprise, empathy and critical thinking aren't easy to replicate and they're not bugs to be patched.
The expert consensus shifted, but the technology limitations didn't magically disappear. We're watching a classic case of narrative whiplash while the underlying reality remains: humans are irreplaceable for the work that actually matters.
Here's what I think is actually happening: We're not headed for human replacement. We're stumbling toward something more like human amplification. The future isn't "person vs. AI", it's "person + AI vs. really complex problems." With more complex analysis, say for legal briefings or medical analysis, what this means is a clear, clean and reliable dataset will be needed - not the all encompassing Gen AI models we’re all using and the vast majority of AI Agents APIs are tapped into. If you use the APIs to Claude, Grok, Gemini or ChatGPT you risk building a spaceship that can only navigate if someone's constantly reminding it which way is up. What that means is reliable AI has two paths. The first is Humans in the Loop and the second is costly proprietary models that are in most instances too expensive and may not make financial sense to update or operate for any non multi-billion dollar org.
Enterprise budgets move like cruise ships, they're large, slow and avoid rough waters (someone's already decided the 2025 dinner menu last year), so we're seeing the "replace everything with AI" experiments now. But 2026? That's when the course correction happens. Not less AI, but a choice between costlier and smarter AI or humans doing more interesting work with AI. Case in point: Mark Zuckerberg just spent last week throwing $50+ billion at AI talent acquisition including $100 million signing bonuses, trying to buy Safe Superintelligence for $32 billion - when that failed tried buying the investment firm backing them, and hiring entire teams from Anthropic and OpenAI. This isn't the behavior of someone who has AGI figured out. This is someone who knows he's nowhere close and is panic-buying every smart person he can find. If Meta, with their infinite resources and the world's best AI teams is this desperate to acquire external talent, when they have talent like Yann LeCun, what does that tell us about how far we actually are from the AI replacing human jobs?
So while the AI experts are busy revising their predictions downward, I'm revising mine upward. The question isn't whether your job is safe from AI. It's whether you're ready to become the kind of person who makes these powerful, occasionally confused machines actually useful, because that's where the real opportunity lies.
This will be featured in my Pandora's Bot Newsletter on July 1st 2025. Enjoying it? Don't forget to subscribe! And if you think your friends or colleagues would enjoy it, please share thepandorasbot.com!