Subvolts take Web agents that see and act autonomously are here, tackling complex tasks with visual intelligence. Nov 18, 2025 - Developers
- Learn about robust techniques for building autonomous web agents using visual grounding and advanced training methods.
- Business
- Understand how proactive AI agents can automate complex web tasks, potentially transforming user interaction and operational efficiency.
This interview provides a solid look at a leading company developing advanced autonomous web agents and their underlying technology.
Proactive Agents for the Web with Devi Parikh - #756
Audio streams from Sam Charrington. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
Devi Parikh of Yutori joins The TWIML AI Podcast to discuss proactive, autonomous web agents. The episode explores technical challenges, the benefits of visually-grounded models over DOM manipulation for web interfaces, and Yutori's training pipeline. They also cover how Yutori's "Scouts" agents use multiple tools and sub-agents for complex queries, the importance of ambient operation, and the path to full web task automation.
Imagine AI that can browse the internet for you, not just by reading code, but by looking at web pages like you do. This episode explains how these "seeing" AI agents work and how they can handle complicated online tasks, making the web easier to use.
This episode offers a deep dive into the future of web interaction through autonomous agents, a key area for frontier AI labs. Listeners will learn about the technical hurdles and innovative solutions, like using screenshots instead of DOM, for building reliable web agents. It's for AI researchers and developers interested in agentic systems and multimodal AI. This work directly informs how companies like Google, OpenAI, and Microsoft are thinking about agent capabilities and user interfaces for their AI products.
- HostSam Charrington
- GuestDevi Parikhco-founder and co-CEO, Yutori
- Guest: Devi Parikh, co-founder and co-CEO of Yutori
- Episode Title: Proactive Agents for the Web with Devi Parikh - #756
- Show: The TWIML AI Podcast
- Length: 56 minutes
- Training Pipeline Evolution: From supervised fine-tuning to rejection sampling and reinforcement learning.
Today, we're joined by Devi Parikh, co-founder and co-CEO of Yutori, to discuss browser use models and a future where we interact with the web through proactive, autonomous agents. We explore the technical challenges of creating reliable web agents, the advantages of visually-grounded models that operate on screenshots rather than the browser’s more brittle document object model, or DOM, and why this counterintuitive choice has proven far more robust and generalizable for handling complex web interfaces. Devi also shares insights into Yutori’s training pipeline, which has evolved from supervised fine-tuning to include rejection sampling and reinforcement learning. Finally, we discuss how Yutori’s “Scouts” agents orchestrate multiple tools and sub-agents to handle complex queries, the importance of background, "ambient" operation for these systems, and what the path looks like from simple monitoring to full task automation on the web.
The complete show notes for this episode can be found at https://twimlai.com/go/756.
- DOM
- Sam Charrington
- Document Object Model, a programming interface for HTML and XML documents that represents the page's structure.
- visually-grounded models
- Devi Parikh
- AI models that understand and process information based on visual input, like images or screenshots.
- rejection sampling
- Devi Parikh
- A technique used in machine learning to generate samples from a target distribution by sampling from a related distribution and rejecting some samples.
- reinforcement learning
- Devi Parikh
- A type of machine learning where an agent learns to make decisions by taking actions in an environment to maximize a reward.
53 min1 h 30Why Jev Is Changing How We Build With AI with Diogo Almeida - #779
WithDiogo Almeida
1 h 081 h 05From Voice Agents to AI Avatars with Alexander Smola - #777
WithAlexander Smola
Questions people ask
- What are proactive web agents?
- These are AI systems designed to interact with the web autonomously, anticipating user needs and performing tasks without constant direct instruction.
- Why use screenshots instead of DOM for web agents?
- Visually-grounded models operating on screenshots are more robust and generalizable for complex web interfaces compared to the brittle Document Object Model (DOM).
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Sam Charrington.