For five years, I was a chump. I chased every shiny new AI ethics framework that came out of a top university or a big tech company. I read the whitepapers, I went to the conferences, I even hired consultants. I thought I was doing everything right. Turns out, I was just spinning my wheels and wasting a ton of money.
This is the story of my biggest professional failure. It’s the story of how I, a two-time founder with a couple of successful exits and over 200 angel investments in companies like Anthropic and OpenAI, got it completely wrong. And it’s the story of the one simple lesson that changed everything.
The Siren Song of the Framework
It started around 2020. AI was exploding, and so was the conversation about its dark side. As someone building and funding AI companies, I felt a deep responsibility to get it right. So, I did what any good Silicon Valley entrepreneur does: I looked for a scalable solution. A system. A framework.
I remember being so excited. We were building a new AI-powered feature at RemoteTeam to help companies analyze employee feedback. The potential for bias was obvious. What if our model favored certain demographics? What if it misinterpreted slang or cultural nuances? The stakes were high.
So we brought in the big guns. We downloaded a 50-page “Responsible AI Toolkit” from a tech giant. It was full of checklists, matrices, and stakeholder maps. It talked about “multi-stakeholder value alignment” and “normative ethical principles.” It felt so… official. So thorough.
We spent the next six months trying to implement it. We had three engineers and a project manager dedicated to it. We held endless meetings to define our “ethical pillars.” We argued for weeks about the precise definition of “fairness.” We built a complex dashboard with a dozen different bias metrics that nobody, not even the data scientists, truly understood.
The result? We delayed the product launch by two quarters. We spent over $200,000 on salaries and resources for an ethics module that was so complicated, it was practically unusable. And the worst part? When we finally did launch, the dashboard was ignored. The product managers just wanted to know if the feature worked. The engineers were burned out. And I had a sinking feeling that for all our effort, we hadn't actually made the product any safer.
The Illusion of Progress
This wasn't a one-time mistake. I saw this pattern repeat itself over and over. At one of my portfolio companies, a startup building AI for loan applications, the founders spent a year trying to create a “provably fair” algorithm. They got so lost in the mathematical weeds that they never actually shipped a product. They ran out of money and the company died.
I was funding failure. I was encouraging it. I was part of the problem. The entire industry was caught in this trap, this illusion of progress. We were so busy creating these elaborate, theoretical castles in the sky that we forgot we were supposed to be building actual houses on the ground. We were performing ethics, not practicing it.
The Single Painful Lesson
My “aha” moment didn’t come in a flash of insight. It was a slow, grinding realization born of watching millions of dollars and thousands of engineering hours go down the drain. It finally hit me during a board meeting for a company that was using AI to detect deepfakes. They were showing me their “Ethical Risk Dashboard,” and it was the same old story. A beautiful, complex interface with lots of red, yellow, and green lights. It looked impressive. But then I asked a simple question: “How does this actually stop a bad actor from using your tech to create a convincing deepfake of a political candidate a week before an election?”
Silence. Blank stares. The CEO started talking about their “risk mitigation scores” and “stakeholder engagement process.” He completely missed the point.
That’s when I knew. The frameworks were a distraction. The checklists were a crutch. The dashboards were just for show. The single most important thing, the only thing that actually moves the needle on responsible AI, is this:
Focus on the data, not the framework.
That’s it. That’s the big secret. It’s not sexy. It’s not going to get you a standing ovation at a TED talk. But it’s the only thing that works. Your AI model is only as good, and only as ethical, as the data you feed it. Period.
From Buzzwords to Reality: What Actually Works
So what does “focusing on the data” actually look like in practice? It’s not about downloading another PDF. It’s about getting your hands dirty. It’s about a fundamental shift in how you build products.
1. Obsess Over Your Training Data: This is where 90% of the ethical problems originate. If your data is biased, your model will be biased. It’s that simple. For our employee feedback tool, we should have spent less time on matrices and more time collecting diverse data. We should have been recording interviews with factory workers in different countries, not just white-collar employees in San Francisco. We should have been paying for high-quality, labeled datasets that represented the global workforce, not just scraping public comments from Reddit.
2. Red Teaming is Non-Negotiable: You need to actively try to break your own models. Hire a team of creative, slightly paranoid people (I call them “professional pessimists”) and pay them to find the edge cases, the loopholes, the ways your AI can be abused. For the deepfake detection company, this would mean hiring people to create deepfakes using their own tools. See how easy it is. See what the failure modes are. Don’t just wait for it to happen in the wild.
3. Build a “Human in the Loop” System: The idea that you can build a fully autonomous, perfectly ethical AI is a fantasy, at least for now. You need human oversight. For the loan application startup, instead of trying to build a “provably fair” algorithm, they should have built a system that flagged borderline applications for a human loan officer to review. The AI’s job should have been to assist the human, not replace them. This isn’t a failure of automation; it’s a recognition of reality. It’s how you ship a product in 2025 instead of going bankrupt in 2026.
The Coming Storm: AI Regulation in 2026
This isn’t just about building better products. It’s about survival. Governments are waking up. The EU AI Act is just the beginning. By 2026, we’re going to see a wave of AI regulation in the United States and around the world. And the regulators aren’t going to be impressed by your fancy ethics framework. They’re not going to care about your stakeholder map.
They’re going to ask for your data logs. They’re going to ask you to prove that your model wasn’t discriminatory. They’re going to run their own tests. And if you can’t show them the receipts—if you can’t demonstrate a rigorous, data-centric approach to safety and fairness—you’re going to be facing massive fines and potentially even criminal liability.
The companies that are just paying lip service to ethics are going to be wiped out. The ones that are actually doing the hard work on the data level are the ones that will survive and thrive.
Stop Performing, Start Building
I wasted five years of my life and a lot of money because I was performing ethics instead of practicing it. I was so focused on looking like I was doing the right thing that I lost sight of what actually mattered.
Don’t make my mistake. Stop downloading frameworks. Stop having endless meetings about your “ethical pillars.” Stop building dashboards that nobody uses.
Instead, go look at your data. Go talk to the people who are collecting it and labeling it. Go hire a team to try and break your model. Put a human in the loop. Do the hard, unglamorous work. It’s the only way to build responsible AI that actually ships. It’s the only thing that will protect you from the coming regulatory storm. And it’s the only thing that will let you sleep at night.
Frequently Asked Questions
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.
What experience informs this perspective?
This perspective comes from over a decade of building companies in Silicon Valley, two successful exits (RemoteTeam to Gusto, MovieLaLa to Gfycat), and investing in 200+ startups including Anthropic, OpenAI, and Scale AI. I write about what I've lived.
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.