I sat on my balcony until dawn, wondering if I had anything left to lose. I’d been laid off, started three companies that failed, and my marriage was over. But the worst part? I was wrong about everything I thought I knew about building products.
In 2021, I was a product manager at a hot RPA company. We had just closed a C+ round, doubled our headcount to 600, and I was living the startup dream. Then OpenAI showed up, and the entire deck reshuffled. Our “rule coverage” moat? Useless. The market stopped caring about how many business rules we had coded. They wanted natural language understanding.
Here’s the thing nobody tells you about tech inflection points: When the venture capital narrative flips, your technology stack is already dead. You just haven’t been told yet.
So how do you spot the next sinking ship before you’re drowning? Three signals. First, when investors stop asking about your feature list and start asking about model capabilities, run. Second, when your top product’s incremental revenue per customer starts declining—your “scale” has become “scale diseconomy.” Third, when your tool makes your users’ jobs harder instead of easier. RPA required business analysts to write rules in scripts. LLMs let them speak normally. That’s not an upgrade—that’s an extinction event.
I missed all three. I paid for it with my job, my savings, and my marriage.
But here’s the rule I learned: Resilience isn’t a feeling. It’s a balance sheet with three accounts.
Cash account: 18 months of living expenses, not 6. Startups don’t die overnight—they bleed out slowly. You need time to decide whether to fold or double down. Trust account: partner only with people you’ve lost money with, not people you’ve made money with. You only see someone’s true character when the numbers are red. Psychological account: you don’t have to be positive. You just have to refuse to believe that this moment is permanent. I sat on that balcony for hours. I didn’t smile. But I told myself: “I’m still at the table.”
After the dust settled, I built a product called HumanizeAI. It solves a ridiculous problem: AI writing detectors flagging your content even when you’ve edited it. I didn’t do market research. I felt the pain myself. In the AI era, you don’t survey users. You become the user.
My validation method was embarrassingly low-tech. First version: a small model running on my laptop. Users sent me their text via WeChat. I ran it, sent it back. Ten users per day, 50 tests total. Only when I saw the data work did I build the actual product. And I chose a lightweight model that runs on a 4-core CPU server. It takes a minute to process 1,000 words. It’s not perfect. But it works. And it costs almost nothing.
Here’s the contrarian truth: In AI products, ‘good enough’ that runs on a potato beats ‘perfect’ that requires a $10/hour API call. Most founders chase the latest model. Business judgment is knowing what ‘good enough’ means in a specific context. That’s not a technical skill. It’s a product skill.
Which brings me to the biggest shift: the product manager’s role is no longer about translating user needs into code. LLMs eliminated that translation layer. Your new job is to define what ‘right’ looks like in an ambiguous situation. That requires business judgment, not technical prowess. The AI can generate the UI, the code, the docs. It cannot decide what constitutes a good outcome for your users.
So what doesn’t change? Five things. People want to save time. People fear being replaced. People pay for certainty. People trust authenticity. People need to feel understood. That list was true five years ago. It will be true five years from now. Technology is the surface. Desire is the ocean. Swim in the ocean, not the surface.
Five years ago, I was a SaaS PM in Shenzhen, surrounded by a team of 600. Today I’m a solo founder coding my own product at a desk in my apartment. Sounds ironic. It’s actually the most logical thing I’ve ever done. I’m an INTJ who craves deep conversation, so consider this article my way of saying: I’m still here. I’m still building. And I’m still at the table.
If you’re navigating the AI wave, or just feeling lost in transition, I can’t give you the right answers. But I can share the holes I fell into. The one thing I know for sure: As long as you’re still at the table, you haven’t lost.
FAQ
Q: How do you know when your current technology stack is becoming obsolete?
A: Watch for three signals: investors stop caring about your feature list and start asking about model capabilities, your top product's incremental revenue per customer declines, and your tool makes users' jobs harder instead of easier. If two of these hit, pivot immediately.
Q: What's the practical implication for AI product managers today?
A: Stop obsessing over the latest model. Your real skill is business judgment: knowing what 'good enough' means in a specific context. The AI can generate code and UI, but it cannot decide what a good outcome looks like for your users. That's your job.
Q: Isn't the author's story just survivorship bias?
A: Possibly. But the principles are testable: 18 months of cash reserves, partnership based on shared losses, and validation at near-zero cost. These are low-risk strategies that reduce the odds of catastrophic failure. The emotional hook is real, but the methods are replicable.