The company's goal is to develop a system for both humans and computers to work together as if they were a single entity by automating tasks within software and workflows based on natural language input.
Adept was founded by some of the most well known AI researchers to create collaborative AI systems that augment human capability in enterprise environments.
Adept has developed exceptional credibility due to its world class founding team from Google/OpenAI, and it has received substantial funding of $415M to achieve unicorn status with its innovative, action oriented AI technology that has a strong enterprise focus.
Adept was founded by three of the most prominent AI researchers: David Luan, who was previously at Google/OpenAI; Ashish Vaswani and Niki Parmar, who are the lead authors of the seminal Transformer paper, and Kelsey Szot, who is also an AI researcher and co-founder.
Adept received a total of $415 million in funding, which included a $65 million seed round from investors such as Greylock, Addition, Root Ventures, and angel investors like Reid Hoffman and Uber CEO Dara Khosrowshahi.
Cloud-based with GPU clusters for model training
Proprietary ACT-1 model architecture extending Transformer technology with multimodal understanding (vision+language), tool use capabilities, and reasoning for software interaction - trained to navigate UIs, comprehend content, and execute complex multi-step workflows
Inferred from technical descriptions and founder backgrounds in large-scale Transformer training at Google/OpenAI
Adept utilizes multimodal AI models that have been trained on trillions of UI interaction tokens to interpret and execute natural language commands to interact with any type of software application. Through computer vision and a proprietary actuation layer it simulates human actions such as clicking and typing without needing to access APIs.
Zapier relies upon pre-defined API connections and pre-defined workflows. Adept can operate on any type of software by utilizing UI understanding and agentic actions; however, this approach will require more computational power than traditional API-driven approaches.
The environment used by Adept to automate workflows is fully sandboxed, therefore users can be given granular permissions and all actions are logged and recorded, then deleted once the task is completed. As such, Adept has the level of security appropriate to be used with highly sensitive workflows and data.
Only enterprise pricing is provided; all sales require a contact with Adept Sales. There are no publicly disclosed pricing tiers, nor are there any free plans offered.
Adept is a tool-agnostic solution, and it will work with any UI-based software (i.e., CRMs, Spreadsheets, PDFs, etc.), and even web apps such as Salesforce and Google Sheets using visual understanding.
Sandbox test environments are provided. There is no free trial option made available to the general public, and all sales require an enterprise sales process.
High levels of success are achieved by Adept in the area of UI localization (93% accuracy), Web Visual Question Answering (VQA) (88.2%), and End-to-End Workflow Planning (88% vs GPT-4’s 59%) as well as in the areas of supply chain, financial services, and healthcare workflows.
No coding, or integration is needed to use Adept. All natural language commands automatically convert into actions across the multiple applications being utilized by the user via the Actuation Layer.
Adept represents the state-of-the-art in Agentic AI, which enables the automation of complex enterprise workflows through UI interactions rather than APIs. This solves a significant gap in software automation, but due to its enterprise-only model and the lack of public pricing and/or API access, Adept is limited in terms of accessibility. Therefore, Adept is best suited for forward-thinking companies who are willing to invest in proprietary AI research and development.
The Adept platform is ideal for large enterprises who are attempting to automate unstructured UI, when other forms of RPA and no-code solutions have failed, but require a significant sales investment.
Adept AI uses proprietary agentic AI (utilizing multimodal models, and UI actuation instead of standard APIs), and achieves superior benchmark results (93% UI localization; 88% workflow completion). Enterprise only business model; no public pricing/ API documentation available. Full stack approach utilizing trillions of UI specific training tokens and customized actuation software differentiates this company from others.
Fair - comprehensive product capability info from official site and analyses, but limited public details on pricing, APIs, customers. Enterprise sales process required for specifics.
Created by former Google Brain/OpenAI researchers with deep experience in transformer architecture. Trillions of UI-specific training tokens utilized; Oracle Cloud Infrastructure utilized; NVIDIA A100 GPU’s utilized.
Industry leading benchmark results: 93% UI localization accuracy, 88.2% web VQA, 88% end-to-end workflows (compared to GPT-4’s 59%). Agentic behavior for enterprise applications.
Full stack approach includes multimodal models, customized actuation DSL (Domain Specific Language) and feedback tools. Visual UI automation differentiates from competitors relying on API’s.
Use cases demonstrated include: Supply Chain (availability of products across multiple locations); Financial Services (extraction of data from PDF documents, updating of internal systems); Healthcare (processing of license requirements, with human-in-the-loop verification).
Translates users’ plain language instructions into actions across multiple applications.
Uses visual cues to navigate and interact with software interfaces as if it were a human performing the task.
Executes complex end-to-end enterprise workflows by breaking them down.
Rapidly learns new workflows through simple human demonstrations and Reinforcement Learning from Human Feedback (RLHF).
Operates across CRM’s, Spreadsheets, Browsers, etc. without needing APIs.
Resistant to UI/UX changes without breaking workflows.
Allows humans to be part of the approval process at the end of critical workflows.
Provides enterprise-grade access control and team collaboration features.
Includes a detailed workflow view of agent thought processes and actions.
Compliant with regulatory requirements when operating across various enterprise systems.
Improves both safety and alignment over time through user corrections.
