Adept site

Adept emerged from stealth on April 26, 2022 with $65 million in funding co-led by Greylock and Addition (with Root Ventures and angels including Andrej Karpathy, Scott Belsky, Howie Liu, Chris Re, and Sarah Meyohas). Its founding team was led by CEO David Luan, who had previously led engineering at OpenAI (helping scale GPT-2/GPT-3-era models) and had run Google's large-models effort at Google Brain, together with Ashish Vaswani and Niki Parmar, two of the co-inventors of the Transformer architecture (the 'Attention Is All You Need' paper), who initially served as Chief Scientist and CTO respectively. Additional co-founders — Augustus Odena, Maxwell Nye, Erich Elsen, and Kelsey Szot (Stanford-trained, Cofounder & President) — rounded out a technically elite founding bench drawn substantially from Google Brain/DeepMind and OpenAI. Adept's thesis was that language models needed to do more than generate text: they needed to act on real software, so the company built 'ACT-1,' a large transformer trained to operate a browser and business applications (e.g., Salesforce, spreadsheets) via natural-language commands, delivered through a Chrome-extension-style overlay.

In March 2023, Adept raised a $350 million Series B led by General Catalyst and co-led by Spark Capital, with participation from Addition, Greylock, Atlassian Ventures, Microsoft, Nvidia, Workday Ventures, and other investors, at a valuation described as 'at least' $1 billion — bringing total funding to roughly $415 million. Over the following year the company shipped a string of research artifacts and products: the open-source Persimmon-8B language model (Sept 2023), the open multimodal Fuyu-8B model built specifically for digital-agent perception (Oct 2023), Adept Experiments (Nov 2023) as a way to let users try its early agent technology, and the larger proprietary Fuyu-Heavy multimodal model (Jan 2024), alongside achieving SOC 2 Type 1 compliance (May 2024). By early-to-mid 2024, however, Adept's original dual ambition — training frontier foundation models while also building enterprise agent products — was reported to require far more capital than the company could efficiently raise, and by some accounts the company faced fundraising difficulty typical of mid-sized foundation-model labs squeezed by hyperscaler-backed competitors.

On June 28, 2024, Adept announced a major strategic shift: Amazon hired co-founder and CEO David Luan along with fellow co-founders Augustus Odena, Maxwell Nye, and Erich Elsen (as well as co-founder Kelsey Szot and additional team members) into a newly formed Amazon AGI organization, with Luan reporting to Rohit Prasad (former Alexa chief, then head of Amazon's AGI effort). Simultaneously, Amazon signed a non-exclusive license for Adept's agent technology, multimodal model family, and associated datasets. Crucially, this was structured as a talent-and-technology license rather than a formal acquisition: Adept continued to exist as an independent company, with no equity stake reported for Amazon and no disclosed purchase price. Commentators and Luan himself later described this as one of the first prominent 'reverse acqui-hires' — the inverse of a typical acqui-hire, where the acquirer takes the founders and licenses the IP rather than buying the company outright.

Business model

B2B enterprise software: AI agents/models sold as a product layer (and later a licensable technology/workflow-automation platform) that automates multi-step tasks across a customer's existing software stack.

Enterprise licensing and product subscriptions for agentic-AI workflow automation (Adept Workflow Language / Adept Agents); prior to the 2024 restructuring, also positioned around foundation-model access.