# Adept AI

Adept AI is a machine learning research and product company founded in 2022 by former Google researchers, focusing on AI agents that automate software workflows and digital tasks.

Adept AI is an artificial intelligence company that develops machine learning models and products designed to automate complex software workflows. Founded in 2022 by a team of former researchers from [Google DeepMind](https://www.wikiprompt.org/wiki/google-deepmind) and other leading AI labs, the company focuses on building "AI agents" that can operate computer interfaces - such as web browsers, spreadsheets, and code editors - to complete tasks that typically require human interaction with digital tools. The company's mission centers on creating a general-purpose assistant that understands user intent and translates it into concrete actions across a wide range of software applications.

The company emerged during a period of rapid advancement in [generative AI](https://www.wikiprompt.org/wiki/generative-ai), particularly in the development of [large language models](https://www.wikiprompt.org/wiki/large-language-model). Adept AI distinguishes itself from many peers by emphasizing action-oriented AI rather than purely conversational or text-generating systems. Its core technology combines natural language understanding with the ability to navigate graphical user interfaces, effectively acting as a digital worker that can perform multi-step tasks such as data entry, report generation, or software testing without requiring custom integrations for each application.

## Founding and Leadership

Adept AI was co-founded by Ashish Kumar, David Luan, and Niki Parmar, all of whom had previously worked at Google and [Google DeepMind](https://www.wikiprompt.org/wiki/google-deepmind). David Luan, who served as the company's first CEO, had earlier been a technical lead at OpenAI, where he worked on early versions of GPT models. Niki Parmar and Ashish Kumar were both researchers on the original [Transformer](https://www.wikiprompt.org/wiki/transformer) architecture team at Google, contributing to foundational work that underpins modern large language models. The founding team also included other notable AI researchers, such as Jakob Uszkoreit, who had co-authored the Transformer paper and later joined Adept as a co-founder.

The company's leadership brought deep expertise in both model architecture and practical deployment. In 2023, Adept hired Brad Lightcap, a former executive at [OpenAI](https://www.wikiprompt.org/wiki/openai), as its chief operating officer, signaling an emphasis on commercial partnerships and enterprise adoption. The team's combined experience in research and product development positioned Adept to tackle the challenging problem of making AI agents reliable and useful in real-world software environments.

## Technology and Approach

Adept AI's technical approach centers on training models that can perceive and act within digital environments. Unlike traditional [machine learning](https://www.wikiprompt.org/wiki/machine-learning) systems that operate on structured data, Adept's models are designed to process screenshots, text, and interface elements to understand the current state of a software application. The company developed a proprietary model family, initially referred to as ACT-1, which was trained on large datasets of human demonstrations of software usage. This training allows the model to predict the next action - such as clicking a button, typing text, or navigating a menu - given the current screen and a user's natural language instruction.

The underlying architecture leverages [Transformer](https://www.wikiprompt.org/wiki/transformer) models, which have become the standard for sequence prediction tasks in AI. Adept adapted this architecture to handle the multimodal input of visual and textual information from computer screens. The models employ techniques such as [multi-head attention](https://www.wikiprompt.org/wiki/multi-head-attention) and [cross-attention](https://www.wikiprompt.org/wiki/cross-attention) to relate the user's request to the relevant parts of the interface. Training involved massive datasets of recorded user interactions, often sourced from anonymized usage data and synthetic demonstrations generated by the company's own tools.

A key innovation in Adept's approach is the use of a "digital twin" environment for training and evaluation. The company built simulated desktop and web environments where models could practice tasks and receive feedback, allowing for reinforcement learning and iterative improvement. This approach, similar to methods used in robotics and game-playing AI, enabled Adept to test its agents in controlled settings before deploying them in real-world scenarios.

## Products and Applications

Adept AI's primary product, initially branded as ACT-1, was a browser extension and web application that allowed users to issue commands in natural language. For example, a user could type "Find all invoices from last month and summarize them in a spreadsheet," and the agent would navigate the relevant web applications, extract the data, and populate a Google Sheets document. The product targeted knowledge workers who spend significant time on repetitive digital tasks, aiming to reduce manual effort and increase productivity.

In addition to the consumer-facing product, Adept pursued enterprise partnerships. The company announced collaborations with major cloud providers, including [Amazon Web Services](https://www.wikiprompt.org/wiki/amazon-web-services) and [Google Cloud](https://www.wikiprompt.org/wiki/google-cloud), to offer its models as part of their AI service catalogs. These partnerships allowed businesses to integrate Adept's agent capabilities into their own workflows without building the underlying infrastructure. Adept also explored vertical applications in fields such as finance, customer support, and software development, where automating multi-step processes could yield substantial efficiency gains.

## Funding and Growth

Adept AI raised significant venture capital funding shortly after its founding. In 2022, the company announced a $65 million seed round led by prominent investors including Greylock Partners and Addition. This was followed in 2023 by a $350 million Series B round, valuing the company at over $1 billion. The funding was used to expand the research team, scale computing infrastructure, and accelerate product development. The company's valuation and investor interest reflected broader market enthusiasm for AI agents as the next frontier in [artificial intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence) applications.

Despite the strong funding, Adept faced challenges common to AI startups, including high compute costs and the difficulty of achieving reliable performance in diverse software environments. The company's burn rate was substantial, and it needed to demonstrate clear customer value to sustain its operations. By 2024, Adept had shifted its focus more toward enterprise solutions and partnerships rather than a standalone consumer product, a strategic pivot that aligned with market demand for practical AI tools.

## Strategic Partnerships and Acquisitions

In 2024, Adept AI entered into a significant partnership with [Amazon Web Services](https://www.wikiprompt.org/wiki/amazon-web-services) (AWS). Under this agreement, AWS obtained a non-exclusive license to Adept's models and technology, and several of Adept's co-founders, including CEO David Luan, joined AWS to lead efforts on AI agents within the cloud platform. This move was seen as a way for Adept to leverage AWS's massive distribution network while continuing to develop its technology. The partnership provided Adept with financial stability and access to AWS's infrastructure, while AWS gained a competitive edge in the emerging field of AI agents.

The deal was not an outright acquisition; Adept remained an independent company, but its core team and technology became closely integrated with AWS's AI initiatives. This arrangement reflected a broader trend in the AI industry where startups with promising research often partner with or are absorbed by larger cloud providers to achieve scale. Adept's technology was also made available through [Microsoft Azure](https://www.wikiprompt.org/wiki/azure) and [Google Cloud](https://www.wikiprompt.org/wiki/google-cloud) in earlier phases, though the AWS partnership became the most prominent.

## Impact and Reception

The AI research community and industry observers have noted Adept AI's contributions to the field of AI agents. The company's work on training models to interact with graphical user interfaces helped popularize the concept of "computer use" as a viable AI capability. Adept's demonstrations, which showed the agent performing tasks like booking flights or filling out forms, attracted significant attention on social media and in tech press, influencing other companies to explore similar directions.

Critics and analysts have pointed out that while Adept's technology is impressive in controlled demos, achieving reliability in the messy, varied world of real software remains a formidable challenge. The company's pivot toward enterprise partnerships was partly a response to these challenges, as businesses are often willing to accept higher error rates in exchange for automation of tedious tasks, provided there are human oversight mechanisms in place. Adept's approach has also sparked discussions about the future of work and the potential for AI to reshape knowledge worker roles.

## Future Directions

As of 2025, Adept AI continues to operate as a research and development entity, with its technology integrated into AWS's agent offerings. The company's roadmap includes improving model accuracy, expanding the range of supported applications, and developing more sophisticated planning and reasoning capabilities. Adept is also exploring ways to make its agents more transparent and controllable, allowing users to understand and intervene in the agent's actions when necessary.

The broader field of AI agents is rapidly evolving, with competitors such as [Anthropic](https://www.wikiprompt.org/wiki/anthropic) and [OpenAI](https://www.wikiprompt.org/wiki/openai) also developing similar capabilities. Adept's unique positioning, with its deep roots in the original Transformer research, gives it a strong technical foundation. However, the company's long-term success will depend on its ability to maintain its research edge while delivering practical value through its cloud partnerships. The next few years will be critical in determining whether Adept's vision of a universal AI assistant for software tasks becomes a mainstream reality.

## See Also

- [Large language model](https://www.wikiprompt.org/wiki/large-language-model)
- [Generative AI](https://www.wikiprompt.org/wiki/generative-ai)
- [Google DeepMind](https://www.wikiprompt.org/wiki/google-deepmind)
- [Amazon Web Services](https://www.wikiprompt.org/wiki/amazon-web-services)
- [Transformer architecture](https://www.wikiprompt.org/wiki/transformer)

---
Source: https://www.wikiprompt.org/wiki/adept
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-09T01:53:01.686717+00:00