For five years, I was a model citizen of the AI ethics world. I read all the papers. I went to the conferences. I could talk your ear off about fairness, accountability, and transparency. I even helped draft a few of those dense, hundred-page AI ethics frameworks for my own companies.
And it was all a complete waste of time.
There, I said it. It feels good to get that off my chest. For years, I thought I was doing the right thing, that I was on the front lines of making AI responsible. Turns out, I was just spinning my wheels, caught in a cycle of academic jargon and corporate virtue signaling. I was so focused on the theory of AI ethics that I completely missed the reality of building actual products.
This is the story of my biggest professional failure. It’s the story of how I, a two-time founder with successful exits and an investor in some of the biggest names in AI, got it so wrong. And it’s the story of the single, painful lesson that changed everything.
The Framework Trap
My journey into the AI ethics rabbit hole started around 2018. The Cambridge Analytica scandal was fresh, and suddenly everyone was talking about the dark side of algorithms. As a founder, I felt a deep responsibility to get this right. So I did what any good Silicon Valley entrepreneur does: I tried to engineer a solution.
At my first startup, RemoteTeam, we were building tools to help manage distributed teams. We started experimenting with AI to predict employee churn and identify burnout risks. The idea was to help managers intervene before it was too late. A noble goal, right?
But then the questions started. What if the algorithm was biased against certain demographics? What if a manager used this information to fire someone preemptively? The “what ifs” piled up. So we decided to create an AI ethics framework. We assembled a committee. We hired consultants. We spent months debating definitions of fairness and writing principles that nobody could disagree with.
Our framework was a masterpiece of corporate speak. It had pillars, and tenets, and guiding principles. It was also completely useless. When my engineers asked me how to implement “algorithmic fairness” in a churn prediction model, the framework offered no real answers. It was a document designed to be admired, not used.
We ended up shipping a watered-down version of the feature that barely used AI at all. We were so afraid of doing the wrong thing that we ended up doing nothing of value. I repeated this pattern at my next company, and I saw it happen in countless startups I advised. Everyone was writing frameworks, but nobody was building better products.
My Wake-Up Call
My “aha” moment didn’t come from a paper or a conference. It came from a conversation with a founder I had invested in. Her company was building AI-powered tools for farmers to optimize crop yields. I asked her about her AI ethics framework. She just laughed.
“We don’t have a framework,” she said. “We have a checklist.”
She explained that every time they developed a new feature, they went through a simple, three-question checklist:
- What is the worst-case scenario? Not the theoretical, sci-fi apocalypse, but the real, tangible harm this feature could cause to a real person.
- How do we know if it’s happening? What are the specific metrics we can track to detect this harm in real-time?
- What is our plan to fix it? If the worst-case scenario starts to happen, what is our immediate, concrete plan of action?
That was it. No hundred-page document. No abstract principles. Just three, brutally practical questions. It was so simple, so obvious, that it felt like a punch to the gut. I had spent five years chasing complexity, and the answer was this simple.
What Actually Works: The 3-Question Checklist
I’ve since adopted this checklist as my own, and I’ve shared it with every founder I invest in. It’s not a magic bullet, but it’s the most effective tool I’ve found for cutting through the noise and building responsible AI that actually ships. Here’s why it works:
1. It Forces You to Be Specific
The problem with most AI ethics frameworks is that they’re too abstract. They talk about “fairness” and “accountability” in broad, sweeping terms. The checklist forces you to think in terms of specific, measurable harm. It’s not about “bias,” it’s about “are we disproportionately denying loans to qualified applicants from a certain zip code?” It’s not about “privacy,” it’s about “could this feature be used to stalk someone?”
2. It Creates a Feedback Loop
The second question is the most important one that most frameworks miss. It’s not enough to anticipate harm; you have to be able to detect it. This forces you to build monitoring and observability into your systems from day one. You can’t just ship a model and hope for the best. You have to be constantly measuring its impact and looking for signs of trouble.
3. It’s Action-Oriented
The third question turns a theoretical discussion into a practical plan. It’s not enough to say you’ll “address” a problem if it arises. You need to have a specific, pre-defined plan of action. Who gets alerted? What’s the first thing they do? How do you roll back the feature? Having a plan in place means you can respond quickly and effectively when things go wrong.
From Theory to Practice
Let’s take the EU AI Act. It’s a massive, complex piece of legislation. You can spend months trying to decipher it. Or, you can use the three-question checklist to get to the heart of what matters.
Instead of trying to become an expert on every clause of the Act, ask yourself: What is the worst-case scenario that the EU AI Act is trying to prevent? How can I build systems to detect if that scenario is happening? And what is my plan to fix it if it does?
This approach doesn’t just apply to regulation. It applies to every aspect of building AI. Take AI safety and alignment. The conversations around these topics can get incredibly esoteric. But at its core, alignment is about making sure your AI is doing what you want it to do. The checklist can help here too. What’s the worst-case scenario if your AI becomes misaligned? How would you know it’s happening? And what’s your plan to regain control?
As an investor in companies like Anthropic, OpenAI, Scale AI, and Hugging Face, I see this practical approach to safety and ethics being far more effective than any top-down framework. The teams that are making real progress are the ones that are constantly asking these hard, practical questions.
Stop Talking, Start Doing
I’m not saying that conversations about AI ethics are worthless. But I am saying that they are no substitute for action. We need to spend less time writing frameworks and more time building better systems. We need to stop admiring the problem and start solving it.
The three-question checklist is not a complete solution. But it’s a start. It’s a way to cut through the noise and focus on what really matters: building AI that is safe, effective, and worthy of our trust.
I wasted five years of my life on AI ethics frameworks. Don’t make the same mistake I did. Stop talking, and start doing. The future of AI depends on it.
Frequently Asked Questions
How has this view evolved over time?
My thinking on most topics has changed significantly over the years. Early in my career, I held many conventional views that experience proved wrong. I try to update my beliefs when the evidence changes.
Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.
How can I apply this thinking to my own situation?
Start by identifying the core principle behind the opinion, not the specific example. Then ask yourself: does this principle apply to my context? If yes, test it in a small, low-risk way before going all in.