🌐 English में देखें
💳 पेड
🇮🇳 हिंदी
Adept
Adept क्या है?
Adept was an AI agent company that built autonomous systems capable of controlling web browsers and software interfaces to complete multi-step enterprise workflows without human intervention at each step. Its flagship research demonstration, ACT-1, was one of the first public proofs that a multimodal large language model could understand a web UI, locate specific elements, and take actions — filling forms, clicking buttons, navigating between pages — in response to a plain-language instruction like "find the Q3 shipment status for account 4821 and update the CRM."
The company was founded in 2022 by David Luan, former VP of Engineering at OpenAI, alongside Niki Parmar and Ashish Vaswani — two of the researchers behind the "Attention Is All You Need" paper that introduced the transformer architecture. Adept raised $415 million from investors including General Catalyst and Spark Capital to develop proprietary agent training data, a multimodal model called Fuyu optimized for web and UI understanding, and a custom actuation layer for executing actions reliably across websites and desktop software.
In June 2024, Amazon acquired Adept in an acqui-hire structure: Luan and most of the technical leadership joined AWS's AGI initiative, where they led development of Amazon's Nova Act agentic technology. Luan departed Amazon in February 2026, with multiple co-founders having exited the company by that point. Adept no longer operates as a standalone commercial product. Its influence, however, is visible across the current browser-agent category — OpenAI Operator, Anthropic Computer Use, and open-source browser-agent frameworks all reflect the research trajectory Adept pioneered with ACT-1.
The company was founded in 2022 by David Luan, former VP of Engineering at OpenAI, alongside Niki Parmar and Ashish Vaswani — two of the researchers behind the "Attention Is All You Need" paper that introduced the transformer architecture. Adept raised $415 million from investors including General Catalyst and Spark Capital to develop proprietary agent training data, a multimodal model called Fuyu optimized for web and UI understanding, and a custom actuation layer for executing actions reliably across websites and desktop software.
In June 2024, Amazon acquired Adept in an acqui-hire structure: Luan and most of the technical leadership joined AWS's AGI initiative, where they led development of Amazon's Nova Act agentic technology. Luan departed Amazon in February 2026, with multiple co-founders having exited the company by that point. Adept no longer operates as a standalone commercial product. Its influence, however, is visible across the current browser-agent category — OpenAI Operator, Anthropic Computer Use, and open-source browser-agent frameworks all reflect the research trajectory Adept pioneered with ACT-1.
संक्षेप में
Adept is an AI Agent that pioneered autonomous browser control for enterprise workflows and was acqui-hired by Amazon in June 2024. Its multimodal models and proprietary web-UI training data were absorbed into AWS's AGI initiative, which produced Amazon's Nova Act agent technology. Adept is no longer available as a commercial product.
मुख्य विशेषताएं
Proprietary Agent Training Data
Adept assembled trillions of tokens of web UI interaction data — clicks, form completions, navigation paths — to train agents that understood how real software behaves rather than relying solely on HTML structure or API documentation. This training corpus was a key differentiator that Amazon licensed as part of the acquisition.
Multimodal Models
The Fuyu model architecture was purpose-built for web and UI understanding, processing screenshots and interface states to localize interactive elements, interpret dynamic content, and plan multi-step action sequences. Fuyu's design influenced the category of models now used in browser-agent frameworks.
Custom Actuation Software
A proprietary domain-specific language and actuation layer handled the translation of model-generated action plans into reliable browser and desktop commands. This separated Adept's approach from prompt-only agent systems, providing more consistent execution across varied UI states and error conditions.
Feedback & Data Collection Tools
Adept built tooling for enterprise customers to submit corrective feedback on agent runs, enabling fine-tuning on domain-specific workflows. This feedback loop was designed to continuously improve agent reliability for high-stakes tasks like invoice processing, compliance checks, and data entry across ERP systems.
फायदे और नुकसान
✅ फायदे
- Time Efficiency — Adept's agentic approach reduced task completion time for multi-system workflows by automating the navigation, data extraction, and form submission steps that knowledge workers repeat hundreds of times per week across enterprise software.
- High Accuracy — The platform's proprietary training on real web UI interactions produced agents that executed tasks more reliably than general-purpose LLM agents prompted to use browsers, because the models were specifically trained on the visual and structural patterns of business software.
- Scalability — Solutions are scalable across various departments, offering enterprise-wide value.
- Ease of Setup — New workflows can be established quickly using natural language instructions, minimizing setup time from months to minutes.
❌ नुकसान
- Initial Learning Curve — Users may need time to familiarize themselves with the advanced features and functionalities.
- Cost Considerations — Enterprise deployment costs were available only through direct contract — no public pricing was ever published — making budget evaluation difficult for organizations that didn't engage directly with Adept's sales team.
- Integration Limitations — Adept is no longer a commercially available standalone product following its Amazon acquisition in June 2024. Organizations seeking active browser-agent or enterprise AI automation capabilities should evaluate current alternatives such as OpenAI Operator, Anthropic Computer Use, or open-source browser-agent frameworks.
विशेषज्ञ की राय
For teams researching the history of the browser-agent category or evaluating current enterprise agentic platforms, Adept's ACT-1 and Fuyu architecture represent foundational reference points — the first credible public demonstrations that multimodal LLMs could control real software interfaces at the task level. Active enterprise deployments should evaluate current alternatives: OpenAI Operator and Anthropic Computer Use continue the technical lineage Adept established.
अक्सर पूछे जाने वाले सवाल
No. Adept was acqui-hired by Amazon in June 2024, with its co-founders and technical team joining AWS's AGI initiative. The standalone Adept product is no longer commercially available. Teams looking for browser-agent or enterprise AI workflow automation should evaluate active alternatives such as OpenAI Operator or Anthropic Computer Use, which continue the technical category Adept pioneered.
Adept built AI agents that controlled web browsers and software interfaces to complete multi-step enterprise tasks using natural language instructions. Its ACT-1 demonstration — one of the first public proofs of a model navigating real web UIs autonomously — and its Fuyu multimodal architecture were foundational contributions to the modern browser-agent category now represented by OpenAI Operator and similar tools.
Adept's ACT-1 research and Fuyu model architecture directly influenced the browser-agent category. Amazon licensed Adept's proprietary agent training data and actuation software, which contributed to the development of Amazon Nova Act. The pattern of training multimodal models on web UI interaction data — rather than pure text — is now a standard approach across leading browser-agent products.