The Advisor Problem Nobody Wants to Discuss
I've been building AI tools for three years now. Every integration I've shipped—whether it's a Hugging Face model wrapped in FastAPI or a Claude API call buried in production—has required someone to make a call about what this technology is actually for. That decision-maker needs to be separate from the person benefiting from the outcome. It's not complicated. It's basic.
Reading about Shivon Zilis testifying in the Musk v. Altman trial made me think about governance structures in a way I hadn't before. Not because of the personal relationships involved—those are their business—but because of what her role actually was. She claims she advised on "Tesla, Neuralink, and OpenAI." Three massive, competing interests. I'm genuinely uncertain whether anyone could navigate that without bias, regardless of personal circumstances.
- She was embedded across multiple organizations simultaneously
- Tesla's AI ambitions directly compete with what OpenAI was building. That's not a minor conflict.
- Neuralink had its own governance problems and regulatory pressure
- The incentive structures alone make impartiality mathematically impossible
What Actually Happened Here
Here's where I need to be honest: I don't fully understand what her actual authority was. The trial record shows she denies being a "chief of staff" but confirms working on "the entire AI portfolio." Those aren't the same thing. One is an executive role with decision-making power. The other could mean she was a coordinator. But the testimony doesn't clarify which. That ambiguity—I'm not sure it's accidental.
When I architect a system, I define permissions clearly. Who can approve? Who can deploy? Who can see the data? Vague titles create vague accountability. And vague accountability in AI governance is where things break.
What interests me more: Zilis testified about romantic history but also about technical decisions spanning years. The romance is tabloid material. The technical decisions are the actual problem. She had access to strategic information across competing organizations. Whether that access was formally titled "chief of staff" or informally understood becomes almost semantical once the access existed.
The Real Product Design Failure
I've made this mistake at smaller scale. Built a feature. Wanted to use it. Realized halfway through shipping that I couldn't be both builder and validator. The bias isn't malicious. It's structural.
OpenAI's early governance had a flaw that nobody wanted to acknowledge: the people closest to Musk were the ones advising on his role in the organization. At some point between 2017 and now, someone should have flagged that the advisor structure was broken. Not because of who Zilis is, but because the design was fundamentally compromised.
I'm building an AI-powered document analysis tool right now. We have an internal review board specifically so I can't approve my own code. Not because I'm untrustworthy. Because I'm human. Same principle applies at scale.
- Separation of concerns. This is Rule One in engineering.
- Conflicts of interest don't resolve themselves through good intentions
- The Anthropic board structure, at least publicly, tried to avoid these traps. Whether they succeeded is another conversation entirely.
- Nobody wants to hear this, but you might need to recuse yourself
What Doesn't Add Up
The trial revealed that Zilis managed Musk's "entire AI portfolio." That means she was making or influencing decisions about Tesla's autonomous driving, Neuralink's regulatory path, and OpenAI's strategic direction. These organizations have fundamentally different missions and competing resources.
I'm not sure how you resolve that even on paper. The conflict isn't between Zilis and anyone else. The conflict is structural. You can't serve three masters who want different things from the same limited resource pool—in this case, Musk's time, capital allocation, and political influence.
The testimony happened. The roles existed. The decisions got made. Now OpenAI is suing, and Zilis becomes the person who had to explain how all of this happened. She testified under oath. The record exists.
What I'm left wondering is whether anyone in the room—lawyers, judges, observers—actually thought about the governance architecture that made this situation inevitable. Because from where I'm sitting, building actual products,
