OpenAI’s board needs to say something

Earlier this year, I ran into Adam D’Angelo at an AI conference in San Francisco’s “Cerebral Valley” neighborhood.

He was there to talk about Poe, the AI bot platform that is his pet project as the CEO of Quora. But I was more interested in his role as a board member of OpenAI, which had just lost two other directors in a matter of a few weeks: Reid Hoffman and Shivon Zilis. Was the board actively looking to fill their spots? I cornered D’Angelo in the halls of the conference venue to ask. He said yes and declined to elaborate.

I couldn’t have expected that D’Angelo would try to oust Altman just eight months later and that D’Angelo would be the only OpenAI board member to survive one of the most dramatic power struggles in Silicon Valley history. Along with Bret Taylor and Larry Summers, he’s now tasked with rebuilding OpenAI’s nonprofit board and bringing in an outside law firm to conduct an internal investigation into Altman’s firing.

I do expect the new three-person board to be formalized very soon. Meanwhile, D’Angelo and the other board members who suddenly fired Altman without notice have yet to share any evidence for doing so. Even with Altman now back as CEO, OpenAI remains in a state of uncomfortable limbo until they elaborate on the vagueness of Altman not being “consistently candid.”

The board’s silence has already created an information vacuum that’s led to the spreading of conspiracy theories and far-flung rumors about what went down. This atmosphere of uncertainty does a disservice not only to OpenAI employees and investors but also to the 2 million developers building on the company’s technology.

After talking with sources with direct knowledge of the attempted coup against Altman and reading OpenAI’s 2016 bylaws, I’ve come to believe that the board was within its rights to fire him the way it did. My understanding is that the board felt it had to move swiftly or risk that Altman would undermine the effort.

Much has already been reported about Altman’s feuding with Helen Toner, one of the four board members who voted to fire him. She recently co-authored a research paper that praised the safety efforts of rival lab Anthropic, which broke off from OpenAI a few years ago. I’ve heard that Altman was quietly trying to convince other board members to vote Toner off and that the board couldn’t agree on who should fill the seats left vacant this year by Hoffman, Zilis, and Will Hurd.

While no one involved expected Altman to go without a fight, I think the board hoped to have more time to explain the detailed reasoning for firing him. The threat of mass employee resignations forced a compromise. I’m told Altman and co-founder Greg Brockman went from needing to be a package deal on the board to both agreeing to come back without their seats. They also initially wanted all the board members who fired Altman to resign, which didn’t happen now that D’Angelo is staying on.

The next shoe to drop will be the outcome of the internal investigation that both sides agreed to as a condition of Altman’s return. I wouldn’t expect Altman to find his way back onto the board until that investigation is wrapped and absolves him of wrongdoing. Regardless of the outcome, there’s still plenty of blood in the water. It’s only going to get messier from here.

It’s not just the stock price: morale inside Meta appears to be on the upswing, too, according to the results of a recently completed employee survey I obtained.

Meta does these “Pulse” surveys twice a year, and they are taken quite seriously inside the company. This most recent survey was done during the last two weeks of October and had a 90 percent participation rate, with over 22,000 employee-submitted comments. Employees are told that their answers are kept anonymous to encourage candidness.

Employees reported an overall 64 percent favorable view of the company, up 13 points from the first half of this year. Confidence in leadership increased 21 points to 47 percent. The biggest spike in positivity was the number of employees who responded positively to a question asking if they would recommend the company as a place to work: a 27-point increase to 74 percent. Meanwhile, the percentage of employees who said they planned to stay at Meta for just one year or less dropped from 15 to 10 percent.

The results are in stark contrast to where Meta was in the first half of last year, when just 39 percent of respondents reported feeling optimistic about the company’s future and 42 percent expressed “confidence in leadership.” Its business was looking much shakier then, and CEO Mark Zuckerberg was in the process of turning up the “intensity” and flattening the org chart.

In an internal post announcing the latest survey results on Wednesday, HR chief Lori Goler said that the submitted comments specifically called out “the launch of Threads and the product announcements at Connect” as making employees “more confident in Meta’s future.” Connect was where the company’s ChatGPT competitor and a slew of AI characters were announced, along with the Quest 3 headset and the new Ray-Ban smart glasses that seem to be well liked.

As I’ve covered before, the development of Threads was unusually fast for a company as large as Meta, and Zuckerberg has been clear that the goal is to kill the platform formerly known as Twitter. Threads usage has certainly died down since its initial debut, but it’s still being updated regularly, with federation on the ActivityPub standard still coming. At the end of the day, everyone loves a good rivalry, even if it likely won’t come to physical combat.

“You shared that the people who work here are your favorite part of working at Meta, and you appreciate the amount of autonomy you have here,” Goler said in her internal post on Wednesday to employees, which also noted that the company “heard feedback around the rigidity of in-person expectations for office workers.”

Zuckerberg instituted a three-day-a-week in-office policy for most employees this year after leaning hard into remote work during the pandemic. I’ve heard that it has been a tough adjustment for many who moved as a result of the previous, remote-friendly policy. “We plan to stay the course on this at this point, and continue to learn about how this is working for our teams,” according to Goler.

Meta is no stranger to scandal. Recent headlines have been homing in on what the company knows about underage users on its apps. There’s still a looming antitrust lawsuit brought by the US government, increasing competition in generative AI, and Apple’s entering the headset wars with the Vision Pro release next year. Even still, the worst seems to be in the rearview mirror for now, and Zuckerberg is winning back the support of his troops.

AI chatbots like ChatGPT have, to date, been fairly impersonal, existing outside of the apps and data that we use every day. A new startup by three ex-Apple employees called Software Applications Incorporated hopes to change that.

The company’s CEO, Ari Weinstein, is a repeat founder, having sold his last startup, the iOS automation app Workflow, to Apple in 2017 alongside co-founder and CTO Conrad Kramer. This time, the two have been joined by Kim Beverett, a 10-year Apple vet who was onstage at this year’s WWDC and previously oversaw product management for various teams, including Safari, Messages, FaceTime, and user privacy.

In their first interview since leaving Apple to start something new, the trio tells me that their focus is on bringing generative AI to the desktop in a way that “pushes operating systems forward.” While they don’t have a product to show off yet, they are prototyping with a variety of large language models, including OpenAI’s GPT and Meta’s Llama 2. The ultimate goal, according to Weinstein, is to recreate “the magic that you felt when you used computers in the ’80s and ’90s.”

“If you turned on an Apple II or an Atari, you’d get this basic console where you could type in basic code as a user and program the computer to do whatever you wanted,” he explains. “Nowadays, it’s sort of the exact opposite. Everybody spends time in very optimized operating systems with pieces of software that are designed to be extremely easy to use but are not flexible.”

An example he gives: “Sometimes you’ve got a browser window open with a schedule on it, and you just want to say, ‘add this to my calendar,’ and somehow, there’s no way to do that… We think that language models and AI give us the ingredients to make a new kind of software that can unlock this fundamental power of computing and make everyday people able to use computers to actually solve their problems.”

The team’s love for early PC nostalgia is evidenced by the Software Applications website, which is literally Mac OS 8 running in a browser tab. Weinstein says they hope to hire up to 10 employees in the coming year, including a designer and some machine learning experts. They’ve already raised $6.5 million in funding from OpenAI’s Altman, Figma CEO Dylan Field, and other notable names in Silicon Valley.

In a world where venture capital money is drying up and there are mass layoffs happening throughout the tech world, it’s a feat to raise that much before you even have a product, though the trio’s past success certainly helps. “I met Ari on the platform formerly known as Twitter when we were both in high school,” Field tells me. “He is one of the most talented people I know, and I would back anything he does.”

Apple doesn’t buy startups that often, and when it does, they rarely continue as distinct products like Weinstein’s last project has. Shortly after Workflow was bought, it was renamed the Shortcuts app, which is now preinstalled on iPhones and Macs. If you have an iPhone 15, you can use Shortcuts to create all kinds of use cases for the new Action Button on the side of the device, such as controlling your home’s smart lights or triggering an action in a third-party app.

Given that Apple appears to have been caught flat-footed in the generative AI race, a natural question is whether Weinstein, Kramer, and Beverett felt they needed to leave the company to build what they’re working on now.

When I ask, Weinstein refutes that notion: “We started this company because we’re so excited about what’s happening in generative AI right now, because we were excited to work together again, and because we love the open-ended, creative environment of a startup.” Beverett, meanwhile, adds that Apple’s in-person work policy became unfeasible after moving farther away from Cupertino during the pandemic.

While a handful of startups such as Rewind are building personalized AI systems for the desktop, none I’ve come across appear to have as wide a vision as that of Software Applications, which suggests they have an open lane for the time being. And with most consumer software companies building primarily for mobile these days, it’s refreshing to see a team focusing on the Mac instead.

“The average interaction on a mobile device is measured in seconds, and the average interaction on a desktop computer is measured in minutes or hours,” Weinstein says. “So saving people time on the desktop is really exciting to us.”

I’ll be back with another issue next week. In the meantime, send me your feedback, story ideas, and theories as to why Mark Cuban is suddenly leaving Shark Tank and selling the Mavs.

Thanks for subscribing.