Adept Unveils Advanced AI Agents with Multimodal Capabilities

Adept has introduced a groundbreaking development in artificial intelligence with its new AI agents, designed to streamline complex web interactions and enterprise automations using the custom-built Adept Workflow Language (AWL). According to Adept.ai, AWL is a powerful language that enables users to create sophisticated workflows with ease.

Key Features of Adept’s AI Agents

Adept’s AI agents are engineered to be reliable, robust, and user-friendly. These agents can translate user intent into actions, managing everything from complex tasks to repetitive chores. The agents are built on a suite of multimodal models that have been trained to understand screens, reason about content, and make plans from the earliest stages of training.

Key characteristics of Adept’s AI agents include:

Reliability: The agents can consistently execute workflows while staying “on rails.”
Robustness: They are resilient to changes in the execution environment, maintaining functionality despite variations.
Ease of Authoring: Instructions for the agents are simple to write and can even be a few lines of natural language.

Adept Workflow Language (AWL)

AWL, a proprietary language developed by Adept, is a syntactic subset of JavaScript ES6. It offers powerful abstractions to define multimodal web interactions. Users write workflows in AWL, which translate directly to model calls. Specific AWL functions allow users to write instructions in natural language, which are then translated into detailed AWL by the model.

AWL functions like click(“Compose”) and act() provide a flexible way to define agent behavior. The former locates elements on the screen and generates function calls to interact with them, while the latter takes natural language inputs to invoke an agent reasoning loop, allowing the agent to make plans and execute tasks dynamically.

Practical Applications

Adept demonstrated how AWL can be used to create workflows for various applications. One example involved viewing a PDF of event attendees, extracting information, and creating a new lead in Hubspot. The same workflow was replicated for Salesforce with minimal changes, showcasing the flexibility and efficiency of AWL.

Another example highlighted the extraction of patient diagnosis notes from a PDF and entering them into an EMR system. The agent could also handle tasks like creating customer records in Stripe from Google Sheets data and managing leads in Salesforce based on inbound emails.

Transformative Potential

Adept’s AI agents have the potential to revolutionize business operations. The ability to create automations in natural language lowers the barrier for adoption, enabling more users to become citizen developers. This can reduce costs, speed up time to value, and increase the appetite for automations across various workspaces.

Moreover, Adept’s agents can handle more complex workflows involving unstructured data, offering increased resilience and robustness. Their inference-time reasoning allows them to adapt to underlying web changes, making them suitable for a wide range of applications.

Future Prospects

Adept is committed to making automations easier for a broader audience, enabling more types of workflows more frequently. The potential for business transformation with Adept’s AI agents is significant, offering simpler, faster automations and the ability to automate more complex tasks.

For more detailed examples of workflows built with AWL, visit the Adept.ai blog.

Image source: Shutterstock