I Studied the Top 1% of Engineers at FAANG
4 Habits That Separate Them From Everyone Else
In 2014, Ian Goodfellow was drinking beer with friends at a Montreal bar when they asked him to solve an impossible problem: teaching computers to create realistic images. They had a plan..
But Goodfellow thought their approach was wrong. Completely wrong.
Instead of arguing, he went home that night and coded a different solution from scratch. By morning, he had invented GANs, generative adversarial networks.
One of the most important breakthroughs in AI history.
Here's what's interesting: Goodfellow's friends weren't stupid. They were smart engineers working on legitimate approaches. But being wrong in the right direction turned out to be infinitely more valuable than being right about incremental improvements.
This pattern shows up everywhere among the top 1% of engineers at companies like Google, Meta, and Tesla. I’ve studied their careers, looking for what separates them from everyone else.
The answer surprised me. It's not that they're always right. It's that they're wrong in ways that compound.
1. They teach, which forces them to be wrong.
Andrej Karpathy could have kept his knowledge to himself. Instead, he created Stanford's first deep learning course, taught 750 students, and now runs a YouTube channel explaining AI to millions.
Teaching is terrifying for smart people. You have to say "I don't know" in front of others. You have to simplify ideas until they're almost certainly incomplete. You have to watch students ask questions that reveal holes in your thinking.
Most engineers avoid this vulnerability. But the best ones seek it out.
There's a psychological trick at work here. When you explain something, you discover what you don't actually understand. When students ask questions, they expose your blind spots. When you simplify, you find the core principles buried under complexity.
Teaching doesn't just help others learn. It helps you unlearn your misconceptions.
2. They move between worlds
Ilya Sutskever went from academic research to Google to co-founding OpenAI to developing ChatGPT to his new company Safe Superintelligence. Karpathy bounced between Stanford, OpenAI, Tesla, and back to OpenAI. Ian Goodfellow worked at Google Brain, OpenAI, Apple, and DeepMind.
Most people see this as career instability. But it's actually idea arbitrage.
Every field develops its own blind spots. Academia gets too theoretical. Industry gets too practical. Big companies get too bureaucratic. Startups get too chaotic.
The people who move between these worlds collect insights that are obvious in one context but revolutionary in another. They become bridges between islands of knowledge.
This requires giving up something most people crave: the comfort of expertise in a single domain. It means being a beginner again and again.
3. They bet on ideas that sound crazy
When Sutskever left Google to co-found OpenAI, most people thought artificial general intelligence was decades away, if possible at all. When LeCun pushed convolutional neural networks in the 1990s, the AI community had largely given up on neural networks.
These weren't just career risks. They were reputation risks.
There's a harsh truth about breakthrough ideas: They usually sound wrong to smart people. If they sounded obviously right, someone would have done them already.
This creates a selection problem. The ideas that will change the world are hiding among ideas that will waste your time. And you can't tell the difference until afterward.
The best engineers have developed a different relationship with being wrong. They see it as tuition for learning what's true.
4. They think in decades
Sutskever isn't just building AI systems. He's planning for a world where those systems might be more powerful than humans. He literally planned doomsday bunkers for his research team.
LeCun isn't just improving today's models. He's arguing that current approaches are fundamentally limited and we need entirely new architectures.
This long-term thinking looks impractical. It is impractical, in the short term.
But here's what happens: When you think in decades, you make different choices. You build things to last instead of building things to ship. You invest in learning that compounds instead of skills that become obsolete.
Most people can't do this because the feedback loops are too long. It takes years to know if you were right. Your quarterly reviews don't reward 10-year thinking.
But the engineers who master this patient approach end up creating the future that everyone else has to adapt to.
There's something almost unfair about how this works.
The traits that make these engineers successful—teaching publicly, changing fields frequently, betting on crazy ideas, thinking impossibly long-term—are exactly the traits that most career advice tells us to avoid.
Stay in your lane. Focus on shipping features. Build a personal brand around being an expert in one thing. Take the safe path.
That advice isn't wrong, exactly. It's just optimized for a different game.
The top 1% aren't trying to be good engineers. They're trying to be right about the future.
And the best way to be right about the future, ironically, is to get comfortable being wrong everywhere else.
Here’s how to turn silence into a superpower: