The Trust Problem We Can't Build Around
I spent last week reading Ronan Farrow's investigation into Sam Altman while simultaneously integrating OpenAI's API into a product I'm launching in June. The juxtaposition felt deliberate, even though it wasn't. Here's the thing nobody wants to say out loud in Slack channels: we're choosing to build on top of infrastructure controlled by people we don't actually trust.
The investigation documents something specific. Not fraud in the legal sense. Not embezzlement. But a pattern where Altman tells different people different things and then experiences no friction about it. He makes promises to board members that contradict assurances to investors. He describes the WilmerHale investigation to some people as thorough and to others as insufficient. The Microsoft executives quoted in the piece aren't alleging crimes—they're describing someone unconstrained by the social cost of being caught lying. I'm not sure that's better or worse than fraud, honestly, but it's definitely different.
What strikes me most is that Farrow spent eighteen months on this and found the technology people themselves already knew. The Ilya memos were circulating. Safety researchers had documented concerns. Everyone in the ecosystem who worked directly with Altman seemed to understand his default mode was telling you what you wanted to hear, then doing something slightly different.
The Product-Market Fit Problem Nobody Wants to Name
Here's where my experience as a developer matters. I've integrated Claude, GPT-4, and Gemini APIs into actual shipping products. I've watched the benchmarks. The improvements are real but incremental. Coding assistance works. Document summarization works. The stuff that fundamentally changes markets—autonomous reasoning, reliable long-horizon planning, actually replacing knowledge workers—none of that exists yet.
- OpenAI runs at a loss
- The cost structure depends on an interpretation of fair use that hasn't been tested in court and probably won't survive if it is
- Everyone is racing toward a finish line that might not exist
Altman built his reputation on saying we're already there. In blog posts. In interviews. In pitches to Middle Eastern sovereign wealth funds. The narrative is: AGI is imminent, therefore move fast, therefore break things, therefore maybe don't document your board investigations. I'm not certain this is intentional deception versus optimism that calcified into habit, but the effect is identical.
What This Means When You're Shipping Code
Building on OpenAI right now feels like standing on someone's shoulders while they're actively swaying. The API is stable. The product works. But the company that owns it is under real scrutiny, might be heading toward an IPO with undisclosed legal exposure, and is led by someone who multiple Microsoft executives compared to Bernie Madoff in seriousness. Not as hyperbole. As a genuine question mark about what happens if the core narrative fails.
I asked a lawyer friend whether embedding OpenAI deep into a product architecture creates liability if something about their transparency or business claims falls apart. He said probably not, but also the legal landscape for AI companies is moving fast and nobody knows. That's the answer I'm working with.
The harder problem is philosophical. When you integrate someone's API, you're implicitly endorsing their judgment about what's safe and what's reckless. If the person making those calls is willing to tell different stakeholders contradictory things about what happened inside the company, at some point that judgment becomes untrustworthy regardless of how good the tokens are.
The Bubble Question Nobody Wants to Ask
Farrow asks at the end whether the bubble will pop. Whether companies will fail. Whether market incentives will eventually force accountability. The honest answer from someone in the room is: maybe, but we're all betting it won't happen before we ship. Every founder I know is working on the assumption that the AI landscape in 2027 will look similar enough to today that bets made now won't completely invert.
But if Altman is wrong about the timeline. If GPT-5 isn't a leap function. If the economics don't work. Then companies built on the premise of rapid deployment and breakneck speed will have to reckon with the fact that they were built on urgency manufactured by someone whose relationship with accuracy is notoriously loose.
I'm still shipping with OpenAI. The product is better for it. But I'm also not holding my breath about what the next eighteen months of investigation will uncover, or what happens when the people bankrolling this stuff start asking harder questions about whether they were actually told the truth.
