Nov 3, 2025 Dec 3, 2025 2025-11-03

In October 2019, I sat down and wrote a blog post predicting ten ways artificial intelligence would disrupt our lives. At the time, I was deep in the trenches, teaching myself AI algorithms alongside blockchain and cryptography. I was convinced I understood where things were heading.

I had this beautifully confident model: algorithms crunching data, tuning parameters, spitting out results. Neural networks finally getting their moment because of big data. Clean. Predictable. Understandable.

Then ChatGPT launched in November 2022, and the entire game changed.

Six years later, I’m not just watching the AI revolution—I’m living inside it. Building an AI-powered task management app. Using Claude to migrate 2,000+ blog posts. Debugging code with AI assistance. Having conversations that would have seemed like science fiction to my 2019 self.

So I went back and reread my predictions. Here’s what I learned.

What I Got Right (And Why)

DeepFakes – I wrote that “AI may turn this post-factual world into a deep-fantasy world, in which any identity will be instantly duplicable.” This aged perfectly, just faster than expected. Voice cloning, photorealistic fake images, sophisticated video deepfakes—all standard now. What I underestimated was the democratization. I thought this would be nation-states and corporations. Instead, teenagers make deepfakes in their bedrooms.

Instant Translation – Predicted StarTrek-style real-time translation. It happened. Google Translate went from comically bad to genuinely useful. I’ve had business conversations where each person speaks their native language and AI handles translation in real-time.

Logistics – Amazon’s AI-driven supply chains. Pandemic-accelerated demand prediction. The invisible revolution I predicted is now table stakes. Products appear on shelves when needed, in quantities matching demand almost perfectly.

Why did I nail these? Because they’re optimization problems with clear training data. The path from “bad but improving” to “good enough” was predictable.

What I Got Wrong (And What That Teaches Us)

Autonomous Driving – I predicted robotaxis replacing Uber drivers and cars reconfigured with inward-facing seats by now. Six years later? Waymo in geofenced areas. Tesla FSD still requiring supervision. The revolution isn’t here yet.

What I missed: the long tail problem. The last 5% of edge cases turned out to be 95% of the difficulty. More importantly, I missed the human factor. People are terrified, not reassured. Every accident goes viral.

Lesson: Technical capability doesn’t equal adoption.

Predictive Trading – I predicted AI would level the playing field, that everyone using similar algorithms would compress profits. The reality? High-frequency trading firms got more sophisticated. The gap widened instead of narrowing.

Lesson: Technology amplifies existing advantages. It rarely levels playing fields.

The Pattern: I overestimated physical-world disruption (autonomous driving) and underestimated information-space disruption (language models). Physical world requires infrastructure, regulations, social acceptance. Information space just requires compute and data.

The Elephant I Missed: Large Language Models

Here’s the uncomfortable truth: I completely missed the biggest AI development of the last six years.

In 2019, I was thinking about narrow, specialized tools. Computer vision for driving. Translation algorithms for language. Each problem had its own specialized solution.

I wrote: “Artificial intelligence is really just a set of algorithms… crunching large amounts of data, while tuning parameters.”

Technically correct. Fundamentally incomplete.

When ChatGPT launched in November 2022, millions of people suddenly had conversations with AI that could write poetry, debug code, explain complex topics, and engage in creative problem-solving. The transformer architecture had existed since 2017. GPT-3 was available since 2020. But something about ChatGPT’s interface unlocked a new understanding.

The shift wasn’t just technical. It was conceptual.

Previous AI was like having specialized tools—hammer, saw, screwdriver. Large language models are like having someone who understands the entire workshop. They don’t just execute tasks. They understand context, maintain conversation, reason across domains.

I’m living inside this transformation. When I built addTaskManager, AI wasn’t a feature—it was a development partner. Code review, architecture decisions, documentation, marketing. When I migrated my blog, Claude understood my writing voice and generated SEO descriptions that matched my style while improving discoverability.

This is the shift: from “AI as tool” to “AI as collaborator.”

What I’d Predict Now (2025 ? 2030)

Having learned humility, here’s what I see:

AI Agents – Forget self-driving cars. The real autonomous revolution will be autonomous working. Not AI that responds to prompts, but agents that plan, execute, and iterate on complex tasks. Tell an AI “launch this marketing campaign” and have it research, create content, schedule posts, monitor engagement, and adjust strategy. By 2030, this will be commonplace.

Knowledge Work Transformation – Junior copywriters will edit AI outputs, not write first drafts. Junior developers will review AI-generated code, not write boilerplate. The valuable skill won’t be doing the work—it will be knowing what to ask for and how to evaluate quality.

AI-Augmented Entrepreneurship – I built addTaskManager in months, not years, because AI handled grunt work. By 2030, one-person companies will compete with established firms. The differentiator won’t be resources. It will be vision and the ability to direct AI effectively.

Personal AI as Second Brain – Not episodic consultations, but persistent AI that understands your goals, style, context. That remembers conversations and learns from decisions. By 2030, your AI partnership will directly impact your productivity and effectiveness.

What This Means for You

Bet on directions, not specific technologies. I could have lost money on autonomous driving stocks. If I’d bet on “AI will transform information processing,” I’d have been positioned perfectly for LLMs. The technologies are unpredictable. The directions are more often than not clear, though.

Hold your models lightly. Remember Georg the knight landing in a supermarket? That’s rapid technological change. I had a confident AI model in 2019. Then LLMs shattered it. The people thriving aren’t those who predicted perfectly—they’re those who adapted quickly.

Build with AI, not around it. Don’t imagine a future where you avoid AI. Focus on taste and judgment. AI can generate, but evaluating quality is increasingly valuable. Move toward work requiring human creativity and decision-making.

The 2019 Version of Me Was Right About the What, Wrong About the How

I knew disruption was coming. I knew it would reshape industries. I was right about the disruption, wrong about the form.

I imagined incremental improvements to narrow AI. We got a paradigm shift to general-purpose language models.

I imagined physical-world automation. We got information-space transformation.

I imagined AI as tool. We got AI as collaborator.

The 2025 version of me is humbler. I’m confident about the direction—AI will continue automating cognitive work. But the biggest changes are often the ones nobody sees coming.

What I know: the winners will be people who embrace AI as a force multiplier. As a solo entrepreneur building AI-powered products, I’m accomplishing things that would have required a team in 2019.

Stay adaptable, keep learning, and build with AI rather than resist it.

Whatever comes next will surprise us. Georg the knight didn’t need to predict the supermarket. He just needed to realize he wasn’t actually in hell.

---

This follows up “10 Ways In Which Artificial Intelligence Is Disrupting Our Lives” from October 2019. Read the original to see how predictions looked before ChatGPT—it’s a fascinating time capsule.