David Luan
Last reviewed
Jun 8, 2026
Sources
16 citations
Review status
Source-backed
Revision
v1 · 1,494 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 8, 2026
Sources
16 citations
Review status
Source-backed
Revision
v1 · 1,494 words
Add missing citations, update stale details, or suggest a clearer explanation.
David Luan is an American artificial intelligence executive and entrepreneur best known as the co-founder and chief executive of Adept AI, a startup that helped pioneer AI agents capable of operating software, and as the leader of Amazon's AGI San Francisco Lab, where his team built the Amazon Nova Act browser agent. Earlier in his career he was vice president of engineering at OpenAI during the development of GPT-2 and GPT-3, and he led large-model research at Google Brain. In February 2026 he announced that he would leave Amazon to start a new venture.[1][2][4]
Luan grew up in Massachusetts and showed an early aptitude for computing, earning a certificate in computer science from Worcester State University at the age of 12.[6] He went on to study at Yale University, where he earned a bachelor's degree in applied mathematics and political science.[6][7]
After Yale, Luan founded Dextro, a video- and image-recognition startup whose application programming interface (API) identified objects and activity in photos, videos, and live streams.[6][7] Dextro was among the first companies to ship a real-time video classification API. In 2017 it was acquired by Axon, the public-safety company formerly known as Taser International, for a reported 7.5 million dollars.[7][8] Luan then served as director of artificial intelligence at Axon, where he ran Axon AI, a research group applying machine learning to body-camera footage, and helped establish an external AI ethics board to advise the company on transparency and accountability in policing.[6][7]
Luan joined OpenAI in December 2017 as a director of engineering and was later promoted to vice president of engineering.[7] In that role he oversaw teams spanning generative modeling, algorithms, reinforcement learning, and language. His tenure coincided with the development and release of two landmark large language models: GPT-2 in 2019 and GPT-3 in 2020. He left OpenAI in September 2020.[7] (Some popular profiles also credit his time at OpenAI with the CLIP and DALL-E models, but both were first announced in January 2021, after his departure.)
Following OpenAI, Luan joined Google as a director within Google Brain, the company's deep-learning research division. He spent roughly 15 months there leading Google's large-model efforts before leaving in late 2021 to start his own company.[2][7]
In early 2022, Luan co-founded Adept AI Labs and became its chief executive. The company launched publicly on April 26, 2022, with a 65 million dollar Series A round led by Addition and Greylock, and its debut emphasized that the founding team included inventors of the transformer, the neural-network architecture underlying modern large language models.[9][10] Adept's nine-person founding team included Ashish Vaswani and Niki Parmar, two of the eight authors of the 2017 paper "Attention Is All You Need" that introduced the transformer, along with researchers Erich Elsen, Augustus Odena, Maxwell Nye, Anmol Gulati, Kelsey Szot, and Fred Bertsch, many of them drawn from Google Brain and DeepMind.[9][10] Vaswani, who served as chief scientist, and Parmar, the chief technology officer, both left Adept later in 2022 to co-found a separate startup, Essential AI.[11]
Adept's stated goal was to build a "teammate for every knowledge worker," a general system that could complete tasks by directly using existing software rather than producing text alone.[10] In September 2022 the company unveiled ACT-1, short for Action Transformer, a large model trained to take actions inside software tools such as a web browser in response to natural-language commands. ACT-1 became an early and influential demonstration of agentic AI, the idea that a model could observe a screen and carry out multi-step digital actions on a user's behalf.[12] In March 2023, Adept raised a 350 million dollar Series B led by General Catalyst and Spark Capital, reaching a valuation of about 1 billion dollars and bringing its total funding to roughly 415 million dollars.[13][14]
In June 2024, Amazon struck a license-and-hire arrangement with Adept: it hired Luan and a majority of Adept's team and licensed the startup's agent technology and models, while Adept continued to operate independently with its remaining staff.[2][5] Such deals, sometimes called reverse acqui-hires, let large technology companies absorb a startup's talent and intellectual property without a formal acquisition. Amazon reached a comparable agreement around the same period with the robotics startup Covariant, bringing in researchers including Pieter Abbeel, and the new lab drew on talent from both deals.[5][15]
Amazon publicly unveiled the resulting research group, the Amazon AGI SF Lab, in December 2024, with Luan as a vice president leading it.[4] Based in San Francisco and initially reporting to Rohit Prasad, Amazon's senior vice president and head of its artificial general intelligence organization, the lab was chartered to build foundational capabilities for AI agents that can take actions in the digital and physical worlds and handle complex workflows using computers, web browsers, and code interpreters.[4]
The lab's first public product, Amazon Nova Act, was announced on March 31, 2025. Nova Act is a model and software development kit (SDK) for building agents that complete tasks in a web browser, such as filling in forms, navigating sites, and placing orders.[15] Amazon reported that Nova Act scored 0.939 on the ScreenSpot Web Text benchmark, ahead of Claude 3.7 Sonnet (0.900) and OpenAI's computer-using agent (0.883).[15] Luan described the lab's ambition as building "an AI system that can help you do anything a human does on a computer."[15]
In August 2025, Luan publicly defended the structure of the Amazon deal amid a wider industry debate over reverse acqui-hires. He said he hoped to be "remembered more as being an AI research innovator rather than a deal structure innovator," called such arrangements "perfectly rational" given the need to assemble both talent and computing power, and argued that solving the remaining research problems on the path to AGI required access to "two-digit billion-dollar" computing clusters that an independent startup could not easily fund.[16]
On Tuesday, February 24, 2026, Luan announced in a post on LinkedIn that he would leave Amazon at the end of that week "to cook up something new." He wrote that "with AGI so close, I decided to spend 100% of my time on teaching AI systems brand new capabilities." His departure was reported by CNBC, Bloomberg, and GeekWire.[1][2][3] The exit followed a late-2025 reorganization that placed Amazon's AGI division under Peter DeSantis, a 27-year company veteran and senior vice president in its cloud unit; Luan said the team would "be in great hands" with DeSantis.[1] News coverage described his exit as the latest in a series of departures of former Adept leaders from Amazon.[2]
Luan did not disclose the details of his next venture when he announced his resignation. As of June 2026 he was no longer with Amazon.[1][2] Throughout his career he has also been a frequent public commentator on AI agents and AGI, discussing the future of autonomous systems and lessons from OpenAI in interviews and podcast appearances.[16]
| Period | Role | Organization |
|---|---|---|
| early 2010s to 2017 | Founder and CEO | Dextro |
| 2017 | Director of AI | Axon |
| 2017 to 2020 | Director, then VP of Engineering | OpenAI |
| 2020 to 2021 | Director, large-model research | Google Brain |
| 2022 to 2024 | Co-founder and CEO | Adept AI |
| 2024 to 2026 | VP and head of the AGI SF Lab | Amazon |