Samir Mody

Samir Mody is an AI engineering leader who helped The Browser Company develop Arc and Dia, an AI-native browser designed to understand personal context and assist with everyday tasks. His central challenge is giving browser-based assistants access to sensitive information and useful actions without surrendering user control.

Mody studied at the University of California, Los Angeles, and worked on Arc’s performance and architecture before leading AI engineering as The Browser Company developed Dia. His responsibilities expanded to model behavior, product experimentation, evaluation, and browser security. He also recruited AI engineers across product development, machine learning, and prompt engineering. Atlassian announced an agreement to acquire The Browser Company in 2025.

How Mody approaches browser-based AI

• Build experimentation into the browser. Mody embedded controls for prompts, models, tools, and personal context directly inside Dia, letting designers, product managers, operations teams, and engineers test ideas against their own daily work. Promising experiments advance through internal use, evaluation, refinement, and release.

• Treat model behavior as product design. He approaches an assistant’s tone, judgment, contextual awareness, and actions as deliberate product decisions shaped by evaluations, model selection, prompting, and feedback. Dia’s model-behavior team emerged after an operations colleague substantially improved its prompts.

• Optimize prompts with GEPA. Mody applies scored tasks, Pareto selection, and model-generated reflection to improve complex language-model systems without fine-tuning, making systematic optimization more accessible to smaller teams.

• Make confirmation a security boundary. Because browsers expose assistants to private data, untrusted webpages, and external actions, Mody treats prompt injection as a product-design problem as well as a technical threat. Dia’s Gmail, Google Calendar, and form-filling features incorporate user confirmation before consequential actions, allowing people to inspect what the assistant proposes.