OpenAI has reached 20 million users for ChatGPT Work, signaling broad adoption and a path toward making AI agents simple, safe, and affordable for white-collar work.

OpenAI’s Thibault Sottiaux outlines ChatGPT Work and AI agents, focusing on making intelligence more affordable and expanding coding agent technology to a broader audience.

ChatGPT Work brings coding agents to the general workforce with a $20 per month Plus plan to maximize accessibility and value.

Product development is framed as a discovery process that expands model capabilities, gathers user feedback, and refines products, as seen with GPT 5.6 improving handling of large document collections, slides, reports, and deep research.

Sottiaux emphasizes balancing frontier capabilities with practical use cases through a product-of-discovery approach that iteratively deploys features and scales productivity, such as GPT 5.6 for large-scale documents and presentations.

OpenAI pursues rapid product iteration driven by model capabilities, highlighted by GPT-5.6’s enhancements and the ongoing evolution of ChatGPT Work through product discovery.

OpenAI is pursuing cost reductions over time, including an 80% price cut with Luna, aiming to deliver greater utility for the same cost as frontier capabilities evolve.

Sottiaux describes ongoing price reductions as a permanent correction, with the goal of more value at the same subscription price and further reductions in frontier capabilities in the future.

There is a clear emphasis on price reduction over time, with model prices decreasing and the expectation that users get more capabilities for the same subscription amount.

Safety and alignment are core commitments, with investments in a safety stack and transparent benchmarking to address concerns about exposing AI to sensitive channels like email and messaging.

Safety and reliability are prioritized through a robust safety stack and public benchmarks to reassure users about access to private data such as emails and messages.

Security and privacy are prerequisites for broad adoption of autonomous AI agents, with trust, risk assessment, and transparent benchmarks guiding access to private information like emails and messages.