The most revealing moment in Sam Altman’s courtroom testimony this week wasn’t about money or governance. It was a single anecdote: Elon Musk, Altman said, floated the idea of handing OpenAI to his children. Altman called it “hair-raising.”

That detail landed in a San Francisco courtroom on Tuesday as Altman took the stand in Musk’s lawsuit against OpenAI. The broader picture Altman painted wasn’t of a wronged co-founder. It was of someone who couldn’t separate his personal control ambitions from an organization explicitly built to prevent any one person from controlling advanced AI.

What Altman Actually Said

Altman’s testimony covered a lot of ground. He described Musk demanding that OpenAI researchers be ranked by their accomplishments and then “take a chainsaw through a bunch” of them, a management style Altman characterized as fundamentally incompatible with how research organizations work. He didn’t deny knowing Musk’s reputation. He just said it didn’t fit.

More pointed was Altman’s observation about Musk’s focus on controlling the company’s initial for-profit structure. Altman, drawing on his time running Y Combinator, noted plainly that “founders who had control usually did not give it up.” The implication was that Musk’s interest in OpenAI was never fully separable from his interest in owning it.

Musk’s legal team pushed back on several fronts, questioning Altman about alleged deception and his web of financial investments. But the optics of the week belonged to Altman. The “kids” line alone will follow this case for years.

For anyone building in or investing around AI, the trial is a reminder of how much of what became the modern AI industry was shaped by a toxic founding dynamic. OpenAI’s early culture, its organizational structure, its decision to pursue a capped-profit model rather than a traditional nonprofit, all of it has Musk’s fingerprints on it, often in ways that weren’t obvious until they became liabilities.

Meanwhile, Enterprise Voice AI Grew Up

While OpenAI’s founding drama plays out in court, the companies building on top of the infrastructure OpenAI helped create are having a very different week.

Vapi, a startup that provides AI voice agent infrastructure for enterprises, just hit a $500 million valuation after winning a competitive evaluation process at Amazon Ring. Ring reportedly tested more than 40 AI voice platforms before choosing Vapi. The company says its enterprise business has grown tenfold since early 2025.

Tenfold in about 15 months. That number deserves to sit on its own for a moment.

What Vapi does is fairly straightforward to describe: it lets companies replace human agents on customer support and sales calls with AI agents. The underlying technology, voice AI that can handle complex, real-time conversations, has been improving fast enough that enterprises are now willing to route actual customer calls through it. Amazon Ring, a consumer hardware and home security brand with millions of customers, isn’t a company that can afford to experiment carelessly with customer experience. Choosing Vapi over 40 alternatives is a serious signal.

The growth pattern here fits what’s happening more broadly across enterprise AI. The “AI for the enterprise” category spent much of 2024 dealing with skepticism about whether the technology actually worked at scale. Those conversations are largely over. The question now is which platforms capture the workflow and which get commoditized. Vapi’s win at Amazon Ring suggests they’re positioning for the former.

The voice channel specifically matters because it’s historically been hard. Text-based AI interfaces, chatbots, email drafting tools, coding assistants, have had cleaner adoption curves because latency and tone errors are less punishing. A slightly awkward sentence in a chat response is forgettable. A slightly awkward moment in a live phone call is not. Companies that crack voice at enterprise scale are solving a meaningfully harder problem, which is probably why Vapi’s valuation reflects a competitive moat rather than just growth rate.

Google Just Made a Startup Category Harder

One more story worth flagging for founders watching the platform risk question closely: Google announced this week that it’s bringing Gemini-powered dictation directly into Gboard. The feature is launching first on Samsung Galaxy and Google Pixel phones.

Gboard has around 1 billion users. Gemini-powered dictation built into the default keyboard is a direct shot at every standalone dictation and voice-to-text app on Android. Companies that built businesses on top of transcription quality, like the startups that positioned themselves against Google’s older, weaker speech recognition, are now competing against the keyboard itself.

This isn’t a new playbook. Google has killed more than a few startup categories by folding the functionality into Android or Chrome. What’s different now is the speed. Gemini-level transcription accuracy moving into the OS layer in 2026 would have seemed like a five-year horizon as recently as 2023.

For investors, the Gboard move is a reminder that “AI-powered version of a thing that already exists” is a fragile thesis unless you have distribution, data, or workflow lock-in that Google can’t replicate in a keyboard update. For founders in transcription, the honest question is whether the differentiation is deep enough to matter when the baseline improves this fast.

The Through Line

Three stories, three different parts of the industry, but they connect. The Musk-Altman trial is a governance story about what happens when personal control ambitions collide with institutional ones. Vapi’s 10x growth is a market story about enterprise AI finally delivering on its promises. Gboard eating dictation is a distribution story about what it means to build close to a platform you don’t control.

All three are, in different ways, about power: who holds it, how it gets exercised, and what it costs when someone decides they deserve more of it than the structure allows.

Altman’s courtroom answer on Musk and the kids was the sharpest version of something OpenAI has been saying for years. They built the organization to prevent exactly this kind of concentration. Musk’s lawsuit is, depending on your read, either a challenge to that claim or proof that it works.