The Record of Dead AI Products
← Essays
Essay · May 15, 2026 · 9 min read

The Quiet Exit

How the acqui-hire became the modal ending for AI startups that almost worked.

On June 28, 2024, Amazon announced it had hired David Luan, Adept's chief executive and cofounder, along with other cofounders and members of Adept's technical staff. Adept had raised about $415 million. The announcement did not call it an acquisition. Adept the company kept its name, named a new chief executive, and licensed its technology to Amazon.

This is a specific shape of death, and the archive is full of it. It is not bankruptcy. It is not a shutdown announced on a Friday afternoon. It is a transaction in which a much larger company writes a check, the founders and key engineers move into new offices, the technology gets absorbed into some internal product line, and the original brand is quietly turned off. The customers get a wind-down email. The Series B investors get a partial return, or no return. The product page redirects somewhere else, and then eventually does not redirect at all.

Call it the acqui-hire that ended in silence. It has become the modal exit for an AI startup that raised real money and built real technology but did not, in the end, build a business that could stand alone. Not the IPO. Not the merger of equals. Not the orderly liquidation. The hire-the-team, kill-the-product transaction, executed by an acquirer that wanted the people and the patents and was indifferent to whatever the startup had been selling.

The structure is consistent enough to describe in steps. A startup raises across several rounds. The technology is good. The market is harder than the deck suggested. Burn outpaces revenue. A larger company, usually one of perhaps eight or ten possible buyers, offers a deal that pays back some fraction of the cap table, hires the founders on multi-year retention packages, and takes a license to the underlying tech. The board agrees because the alternative is worse. The product is sunset on a timeline of three to twelve months.

Adept is the cleanest recent example. But the pattern is older. In January 2021, ServiceNow bought Element AI, the Montreal lab that had raised about $257 million from the Canadian government, Intel, Nvidia, and Microsoft. The headline price was roughly $230 million. ServiceNow kept the researchers, folded the AI work into its platform, and wound down Element's existing customer contracts.

Vicarious is the same story with different patrons. The company had raised about $250 million from Jeff Bezos, Elon Musk, Mark Zuckerberg, and Samsung, among others, to build general-purpose robot intelligence. In April 2022, Alphabet's Intrinsic announced it had acquired the Vicarious team. The Vicarious brand stopped appearing in press releases. The founders kept working on the same problems inside a different building.

The autonomy sector ran this play repeatedly. Apple acqui-hired Drive.ai in June 2019, picking up engineers for Project Titan after Drive had raised $77 million and tried to run a self-driving shuttle pilot in Frisco, Texas. The brand was retired within weeks. In March 2021, Cruise acquired Voyage Auto, which had raised about $52 million to run autonomous shuttles in retirement communities. The Voyage product was folded into Cruise's roadmap and the name went away. Magna bought what remained of Optimus Ride in January 2022, taking the team and letting the company cease operations.

Otto sits at the strange edge of this pattern. Uber acquired Otto in August 2016 for about $680 million in stock and shut the self-driving trucks division in July 2018. The Otto brand had been retired before then.

Voice and audio went the same way, faster. Spotify bought Sonantic in June 2022 for a reported $93 million. Sonantic had built a synthetic voice tool used by film studios. The standalone product was killed within a year and the team was redirected into Spotify's DJ feature. Descript acqui-hired Lyrebird in September 2019. The Lyrebird brand disappeared and the voice cloning technology resurfaced as Overdub inside Descript. In both cases the underlying research kept running. The companies did not.

Enterprise AI followed the same arc on a longer timeline. IBM had spent years positioning Watson as the future of healthcare. In January 2022, IBM sold the Watson Health data and analytics assets to Francisco Partners, which rolled them into a new company called Merative. The Watson Health brand was retired, and the data and analytics products now belong to a private equity portfolio company most people have never heard of.

The looser examples are looser because the product is still on. Microsoft's March 2024 arrangement with Inflection took a license to its technology and hired Mustafa Suleyman, Karen Simonyan, and much of the technical staff, while Inflection kept Pi online under a new chief executive. Google's August 2024 deal with Character.AI followed a similar shape: a licensing payment, Noam Shazeer and Daniel De Freitas back at Google, and the consumer product still running.

What survives these transactions is worth listing precisely. The founders' careers survive, usually with a step up in title and a significantly larger compensation package. The technology survives, often as a feature inside the acquirer's existing product. The patents survive and get assigned. Some fraction of the engineering team survives, the fraction varying with the acquirer's needs. The research papers keep being cited.

What does not survive is also worth listing. The product does not survive. The brand does not survive. The customer relationships do not survive, except as a list of accounts the acquirer may or may not try to convert. The Series B and later preferred stock often does not survive in any meaningful financial sense, because the headline price gets eaten by liquidation preferences and retention packages negotiated for the founders. Common stock held by early employees who have already left tends to be worth zero or close to it.

Mindstrong is the small version of this. The mental health startup had raised about $160 million before SonderMind bought the remaining technology in March 2023 and the company shut down.

There is a question buried in all of this about who actually gets paid when an AI startup ends. The numbers, where they have been reported, suggest a consistent answer. The founders get paid. The acquirer's shareholders are mostly indifferent because the checks are small relative to market cap. The startup's preferred investors get a partial return, often less than they put in. Most of the equity stack below the founders gets very little. The customers get a wind-down. The product gets a sunset email and a 404.

The Adept page now redirects. The Element AI site is gone. Vicarious.com points to Intrinsic. Drive.ai's domain has changed hands. Voyage's blog is no longer accessible. Sonantic.com redirects to a Spotify subpage. Lyrebird.ai redirects to Descript. Otto's truck routes are someone else's data. Watson Health is Merative. Optimus Ride's last blog post is from December 2021. The companies are gone in the way that companies in this era go: quietly, after the cash and the talent have already moved.

Referenced in this essay

More essays

The Newsletter

New entries and essays by email.

Occasional dispatches when something dies. No spam. Unsubscribe anytime.