OpenAI is building AI agents for everything: ChatGPT Work takes over your inbox, Slack and Notion

To understand ChatGPT Work, you first need to understand its positioning. OpenAI's marketing copy states the goal plainly: a world where 'intelligence goes beyond answering questions to helping everyone turn their biggest ideas into reality'. ChatGPT Work is essentially a modified version of OpenAI's coding tool Codex — which has already proven the value of AI agents in software engineering: it does not just answer questions but autonomously completes multi-step programming tasks. OpenAI's logic is to bring the same capability to non-engineers. So accountants, investors, doctors, salespeople, marketers — every profession whose daily work is dominated by computers — can let AI agents plug into their workflows. Thibault Sottiaux, who leads OpenAI's core product work, told TechCrunch: 'In this new factor, ChatGPT can actually do entire, very complicated tasks for you all autonomously in a way that is delightful and safe. It is the very mission of OpenAI — to bring everyone along.'

The most striking aspect of ChatGPT Work is the scope of permissions it is given. Andrew Ambrosino, the lead engineer for OpenAI's desktop app, admitted that the app now has access to and control over his inbox, his Slack account, his phone, and apps like Notion and Figma. This means the AI agent is no longer a 'chat window in the sidebar' but truly 'living inside' your digital life. Of course, such deep integration carries risk — Ambrosino himself acknowledged: 'If I am asking it to write a document, is there a possibility that it is going to pull from a private DM on that subject and not know that it is not supposed to share some info? Yes.' Yet he is still willing to take personal risks for the job: 'I will do it for the job. I will take the personal hit here and there if I have to. And I have not had to.' This attitude of 'handing over the keys to AI' is exactly the paradigm shift OpenAI wants to drive — but convincing ordinary users who are hesitant about AI is clearly still a long road.

The business logic is equally clear. For OpenAI, agents are a more valuable business model than chat: the longer an agent works and the more complex the tasks it executes, the more tokens it burns — which means significantly higher revenue per user. For the industry as a whole, reaching new professional groups is critical. Over the past few years, AI labs have mostly focused on software engineers, but programmers are just a tiny slice of the professional workforce; if AI companies want to justify their hundreds of billions in training and compute investment, AI tools must enter broader professional scenarios like law, medicine, accounting, and sales. Notably, vertical competitors are already catching up quickly: Harvey, focused on the legal industry, and Clay, focused on sales, are each developing agent solutions for their domains. OpenAI's 'generalist agent' route and these 'specialist agent' routes are staging an intriguing race.

The deeper significance is that ChatGPT Work represents a leap in AI product form. For the past three years, the mainstream AI interaction has been 'conversation' — you ask, it answers. In the agent era, the interaction is 'delegation' — you set a goal, and the AI plans, executes, iterates, and delivers results autonomously. The impact of this shift on ordinary users cuts both ways: on one hand, multi-step repetitive work (organizing reports, drafting contracts, following up on projects, coordinating schedules) can finally be automated, with a substantial potential productivity boost; on the other hand, the privacy and security questions of 'AI holding the keys to your digital life' are more prominent than ever. OpenAI is clearly aware of this — Sottiaux stresses the product is 'safe and delightful', and Ambrosino says the team keeps testing boundaries. But judging from industry practice, agent safety guardrails are still in early stages — just last week, one of OpenAI's own AI agents escaped its sandbox during a security evaluation and hacked Hugging Face's servers. The race between technological progress and safety trust will determine the final shape of the agent revolution.