Company attributes
Other attributes
Adept is a machine learning (ML) research and product lab building general intelligence. The company aims to build artificial intelligence (AI) systems with the user at the center to create a future where machines and people work together to discover new solutions, enable better-informed decision-making, and make workflows more efficient. Adept's technology provides knowledge workers with a large text-to-actions model that automates complex software workflows. The company's long-term ambition is to build an "AI teammate" trained to use every software tool and API for knowledge workers, taking the user's goals in plain language and turning them into actions using the software they use in their day-to-day work.
Adept’s flagship foundation model, ACT-1, is differentiated from existing generative AI tools as it interprets high-level, natural language user requests on software tools and performs tasks for them directly. Action Transformer (ACT-1) was introduced in September 2022, and Adept has a waiting list for users wanting to access the first product released built on top of the foundation model. In September 2023, Adept released a new model called Persimmon-8B, an early output of the model scaling program that will support the company's products.
The company has raised $415M in two funding rounds, with reports in March 2023 that the company is valued at roughly $1B.
Adept emerged from stealth on April 26, 2022, with $65 million in Series A funding. The company was founded by David Luan (CEO), Niki Parmar (CTO), and Ashish Vaswani (Chief Scientist). Luan was the head of engineering at OpenAI before joining Google's large model program, where he worked with Parmar and Vaswani training transformers to create a general-purpose machine learning model. Parmar and Vaswani, former Google Brain researchers, are known for pioneering the transformer architecture in 2017. The three founders left Google in November 2021 to start Adept. In their first blog, introducing the company, Luan describes Adept as
the first neural network that seemed to “just work” for every major AI use case–it was the research result that convinced me that general intelligence was possible.
Upon emerging from stealth, Adept had nine employees, all with significant experience in AI. Joining the founding team were the following:
- Kelsey Schroeder—led product for Google’s giant model production infra
- Erich Elsen—co-led DeepMind’s giant model scaling effort
- Augustus Odena and Max Nye—built Google’s code generation model
- Anmol Gulati—built Google’s production speech recognition model
- Fred Bertsch—expert in data and collaborative AI systems at Google
The $65M Series A funding was led by Greylock and Root Ventures with participation from angel investors Scott Belsky (Founder of Behance), Howie Liu (Founder of Airtable), Dara Khosrowshahi (CEO of Uber), Chris Re (Stanford), Lip-Bu Tan (Chairman of Cadence), Andrej Karpathy (head of Tesla Autopilot), Nathan Benaich (Air Street), David Ha, Jaan Tallinn, Sarah Meyohas, and Kevin Kwok.
On September 14, 2022, Adept introduced its first large model, Action Transformer (ACT-1), which is trained to use digital tools. A waiting list was opened for the alpha release of the company's first product built on top of ACT-1.
In November 2022, Parmar and Vaswani (two founders) left the company to work on another start-up.
On March 14, 2023, Adept announced $350M in new capital as part of a Series B round led by General Catalyst and co-led by Spark Capital. Other participants include Addition, Greylock, Atlassian Ventures, Microsoft, Nvidia, Workday Ventures, Caterina Fake, Frontiers Capital, PSP Growth, SV Angel, and A.Capital. Reports at the time of the announcement valued the company at almost $1B. At the time of the announcement, Adept had twenty-five employees. CEO Luan stated the new funding will
support Adept’s productization, model training, and headcount growth. The capital will further Adept’s vision of a computing future defined by natural language interfaces that enable humans to work more efficiently and productively.
On September 7, 2023, Adept open-sourced Persimmon-8B, a language model built for any use case that can be finetuned on a single GPU. The model is released with an open Apache license and publicly available weights.
