Blog

Thoughts on AI agents, reinforcement learning, and the journey of building intelligent systems.

Benchmark results comparing Amazon Nova Act with Claude Haiku 4.5 and Claude Sonnet 4.5 across REAL Bench, ScreenSpot, and WorkArena

December 2, 2025 Amazon Science

Building AI agents that actually work: Introducing Nova Act as a service

Nova Act's path from research preview to a production AWS service, with public results across REAL Bench, WorkArena, and ScreenSpot.

Product Launch Reinforcement Learning Computer Use

Read the launch story

More writing

April 16, 2026 Amazon Science

A Practical Recipe for Training Computer-Use Agents with RL

Our approach to training computer-use agents through reinforcement learning — addressing data, reasoning, algorithms, and infrastructure challenges for agents that navigate real websites.

Reinforcement Learning Computer Use AI Agents

January 30, 2026

Welcome to My Blog

An introduction to what I'll be writing about — AI agents, reinforcement learning, and the path to human-like intelligence.

AI Introduction Research