Code 2025
From Arc to Dia: Lessons learned building AI Browsers – Samir Mody, The Browser Company of New York
Overview
This talk details the journey of The Browser Company in evolving from their initial browser, Arc, to their AI-native browser, Dia. It emphasizes the critical lessons learned in building AI-powered products, focusing on optimizing iteration speed, treating model behavior as a craft, and integrating AI security as a fundamental aspect of product development. The core thesis is that embracing technological shifts with conviction is essential for company-wide evolution, not just product updates.
Who should watch
- AI Engineers and Builders
- Product Managers
- Those interested in AI-native product development
- Individuals focused on browser technology and AI integration
- Teams looking to improve iteration speed and product quality
Key takeaways
- Optimizing tools and processes for faster iteration is crucial, involving investments in AI feature prototyping, building and running evaluations, collecting data, and automation for continuous improvement.
- Internal tools, including prompt editors and context management, were integrated directly into the product to significantly increase ideation and iteration speed, broadening access to more team members.
- Mechanisms like Jeba, based on reflective prompt mutation and PA selection, enable efficient prompt optimization without complex fine-tuning, allowing for text tuning rather than weight adjustments.
- Model behavior is treated as a craft, encompassing behavior design, data collection for measurement and training, and model steering through prompting and parameter definition, with an iterative build-refine-evaluate-ship cycle.
- AI security, particularly prompt injections, is an emergent property of product building, requiring a blend of technical strategies and user experience design, such as user confirmation steps for sensitive actions like autofill and event scheduling.
- The evolution to an AI-native product required a company-wide shift, impacting hiring, training, communication, and collaboration, underscoring the need to embrace technological change with conviction.
- The formation of the model behavior team highlights that individuals from various roles, not just engineers, can significantly contribute to shaping AI product capabilities.
Notable quotes
*The browser is one of the most important pieces of software in your life and it wasn't getting the attention it deserved.*
*AI is going to transform how people use the internet and in turn fundamentally change the browser itself.*
*When you recognize that technology shift, you have to embrace it. And you have to embrace it with conviction.*
Unofficial community note. Prefer the recording for nuance.