On November 17, 2023, OpenAI’s board fired Sam Altman as CEO with almost no warning and little public explanation. Two days later, the board called Emmett Shear — who had resigned as Twitch’s CEO eight months earlier to spend time with his newborn son — and offered him the job of running the company at the center of the generative-AI boom. He said yes the same day. By November 21, the board had reversed course, Altman was back, and Shear’s tenure as OpenAI’s chief executive was over before it had really started: roughly two days, most of them spent, by his own account, trying to figure out why Altman had been removed in the first place.

Eighteen months later, Shear used what that weekend taught him to start a company built on a bet that the entire AI industry’s approach to controlling the thing it’s building might be pointed the wrong way.

A once-in-a-lifetime opportunity, accepted the same day

Shear’s route to that phone call ran through two companies he’d already built. He co-founded Justin.tv in 2007 with Justin Kan, Michael Seibel and Kyle Vogt — originally a webcam livestream of Kan’s daily life — and became CEO when its gaming offshoot, Twitch, spun out as its own company in 2011. Amazon bought Twitch for roughly $970 million in cash in 2014, one of the largest acquisitions in Amazon’s history at the time. Shear stayed on as CEO for nearly another decade before stepping down in March 2023, saying he wanted to spend time with his family after the birth of his son.

That’s where he was — by his own description “happily avoiding full-time employment” as a part-time partner at Y Combinator — when OpenAI’s board called. “Today I got a call inviting me to consider a once-in-a-lifetime opportunity: to become the interim CEO of OpenAI,” he wrote on X after accepting. “After consulting with my family and reflecting on it for just a few hours, I accepted.” He framed the decision as a sense of obligation rather than ambition: “I took this job because I believe that OpenAI is one of the most important companies currently in existence… Ultimately I felt that I had a duty to help if I could.”

Two days spent investigating his own board

What Shear actually did with the job was try to find out why it existed. “Before I took the job, I checked on the reasoning behind the change,” he wrote. “The board did not remove Sam over any specific disagreement on safety, their reasoning was completely different from that.” He said he’d push for an outside investigation into “the entire process leading up to this point” and for governance changes if the review called for them, and he was blunt about the damage already done: “it’s clear that the process and communications around Sam’s removal has been handled very badly, which has seriously damaged our trust.” He never got to run that investigation. By November 21, employee pressure and investor alarm had produced a deal to reinstate Altman, and Shear’s brief tenure ended with the same speed it began.

From running the company to questioning its method

Shear didn’t found a new company immediately. It took until March 2025, when he launched Softmax with two co-founders: Adam Goldstein, who’d previously founded flight-search startup Hipmunk and then spent years researching bioelectricity and active-inference agents in a biology lab at Tufts, and David Bloomin, an infrastructure engineer with stints at Google, Facebook and Asana who had co-founded the Plurality Institute. The startup raised an undisclosed seed round from Andreessen Horowitz, Lionheart Ventures and Plural, according to Ashlee Vance’s Core Memory, which broke the story.

The premise, as Shear laid it out to Core Memory, is that the dominant approach to AI safety — training a single powerful model and then trying to constrain or supervise it from outside — is a dead end. “We’re sort of at this point where we seem quite happy enslaving the wish-granting God machine,” he said, describing the industry’s default posture toward increasingly capable models. His alternative, which Softmax calls “organic alignment,” looks to biology instead of hierarchy: multicellular organisms and ant colonies function because many individual units find their own role inside a larger whole without a central controller dictating every action. “A bunch of peers come together and find their role in a greater whole together where they maintain their individual identity,” Shear said. “We need a real source of alignment. No one has a secret plan.”

What that actually looks like, so far

In practice, Softmax’s public work so far is early-stage reinforcement-learning research rather than a shipped product — small models, from hundreds of thousands to a few million parameters, trained in virtual environments designed to study how cooperative behavior emerges among multiple agents. The company’s blog, softmax.com/blog, has been publishing that research as it goes, including an April 2026 piece on what it calls teaching AI systems to “supercooperate.” There’s no announced product timeline, and Shear himself has been candid that this is a research bet, not a go-to-market plan: the team he’s built out since 2025 — researchers with backgrounds spanning cognitive architecture PhDs, reinforcement-learning labs and infrastructure roles at Asana and Apple, according to Softmax’s own team page — has grown well past the ten people reported at launch, but the company still isn’t saying what, if anything, it will sell.

What founders should actually take from it

The easy version of this story is a founder pattern-matching his way from one crisis into a second act — Shear had already built and sold one company, briefly ran the industry’s most scrutinized one, and used what he saw there to start a third. But the more useful detail is what he chose not to do with that platform. Two days inside OpenAI’s board crisis would have made Shear one of the most recognizable names available to any AI lab, venture fund or media company that wanted a marquee hire; instead he spent a year and a half largely out of the spotlight before committing to a research thesis that, on his own telling, has no clear product and no near-term way to prove itself right. Softmax’s approach could turn out to be wrong, or simply too slow to matter against labs racing to ship more capable models under the “control” paradigm Shear is betting against. What the two days at OpenAI bought him wasn’t an answer — by his own account, he never got to finish the investigation he opened the job to run — but a specific, close-up view of what he considers the wrong problem being solved the wrong way, and the standing to spend real time and other people’s money testing an alternative instead of chasing the obvious next job.

This is an editorial profile assembled from public reporting, primary announcements and Softmax’s own site. Sources: Fox Business — What to know about OpenAI’s new interim CEO Emmett Shear, Bloomberg — OpenAI’s Interim CEO to Probe ‘Badly Handled’ Altman Exit, CNBC — OpenAI is hiring former Twitch CEO Emmett Shear to run company, NBC News — OpenAI CEO Emmett Shear’s social media history, Core Memory (Ashlee Vance) — Exclusive: Emmett Shear Is Back With a New Company and A Lot of Alignment, Amazon press center — Amazon.com to Acquire Twitch, GeekWire — Amazon’s $970M acquisition of Twitch, Wikipedia — Emmett Shear, Softmax — Team, Softmax — Writing.