OpenAI’s head of product, Thibault Sottiaux, outlined in a recent interview how the company is extending its agent-based AI tools beyond coders and data scientists, aiming for broad adoption with its new product, ChatGPT Work. The platform is designed for white-collar users who want to automate complex tasks across email, document review, report generation, and more. The goal: make these capabilities accessible under the Plus plan at $20/month. The utility, Sottiaux said, is already high, and the ambition is to bring it to as many people as possible.
ChatGPT Work is built on top of OpenAI’s core stack—ChatGPT classic, Codex, the API infrastructure, and agent frameworks—but packaging them for general users. The idea is to wrap these powerful AI agents in an interface that’s safe, intuitive, and compatible with mobile and web platforms. The transition involves not just adding features but changing how users interact, reducing friction so the model takes lead where appropriate, and offering simplicity even as capability deepens.
Capabilities and public readiness
Agents are now capable of more sophisticated work: processing large document sets, producing polished slide decks and reports, and conducting deep research—tasks professionals normally spend hours on. This attaches to GPT-5.6’s recent advances. As these abilities improved, user feedback and real-world deployments sharpened focus on where the models excel, and OpenAI is leaning into those areas for broader release.
Sottiaux believes “the world seems to be ready” for this shift. Proof comes via adoption metrics: ChatGPT reached 20 million users, which he described as an indicator that people are comfortable letting the model drive rather than always requiring manual control. This signals growing trust in AI agents to work semi-autonomously.
Pricing, safety, and evolution of experience
To address cost concerns, OpenAI has imposed a major “price correction” on its frontier models. Most notably, on July 30, the API price for GPT-5.6 Luna dropped 80% (both input and output costs), and the mid-tier Terra fell by 20%, while the flagship Sol model remained unchanged. This makes cutting-edge capabilities cheaper and more available.
On safety and privacy, Sottiaux emphasized that OpenAI’s investments are ongoing: picking secure aligned models, using a multilayered safety stack, publishing benchmark results, and controlling access to new features like letting models reach into sensitive data sources (e-mail, messages) only once robust safeguards are in place.
Voice interaction has also moved forward. The “Live” Voice experience is growing, enabling more natural conversations with pauses, interruptions, and overlapping speech—in effect, letting users talk and listen more human-like. Visual responses, memory, search features are part of this evolving voice mode across platforms.
This isn’t about giving people flashy tools—it’s about expanding who uses AI, reducing the cost of intelligent computation, and balancing autonomy with safety. OpenAI is betting that simpler interfaces, lower price points, and more natural interaction (voice, agents) will shift AI from specialist domains into everyday work.
Why it matters: This moment marks a pivot from AI tools built for experts to systems crafted for everyone who deals with e-mail, documents, schedules, and decision fatigue. It raises new questions: how do we ensure trust and privacy when agents are very capable? How will costs keep coming down across models? And how will the competitive landscape adjust as usability becomes as important as raw intelligence? For users and businesses alike, the upcoming period will test whether AI can really earn its place in the workflow beyond hype, and not just for those who are already deeply technical.
