OpenAI has announced "ChatGPT Work." This is more than a tool that answers questions in a chat. It is an agent, meaning an AI that works on its own. It can take action across your apps and files, stay on one project for hours if needed, and turn a goal into finished work. This is the big difference. In the past, ChatGPT was a smart advisor. Now its role has moved up a level, to a partner you can hand real tasks to.

So how does our work change? Tasks with many small steps, like simple research, first drafts, or copying data, are the easiest to hand over to an agent. What remains for humans is setting the goal, judging the result, and taking responsibility. As AI handles more of the doing, human value shifts toward deciding what to make. Whether you can become the one who gives the instructions may be the key difference going forward.

Behind this shift is a new model, "GPT-5.6," announced at the same time. What matters is not the raw power itself, but the stronger performance per dollar. When the same task can be done more cheaply, companies find it easier to try AI. Because agents run for long periods, they use more computing power, so cost efficiency is the lifeline of wider use. Just as with cloud and smartphones, the lower the unit price, the faster adoption tends to grow.

AI is entering not only the making side of work but also the defense side. Microsoft explained in a blog post that it has started using AI to find vulnerabilities, meaning weak points in software, earlier. As a result, each monthly update is likely to include more fixes. However, attackers are also using AI, so even amateurs can more easily try an attack. Both offense and defense are speeding up with AI. For readers, not delaying updates has become an even closer form of risk management.

It is not all good news. The New York Times claims that OpenAI hid evidence that could hurt it in a copyright trial, and it filed a motion asking for sanctions. As agents spread, the rules about what AI learns from and what it outputs matter more and more. This is a structural risk that could shake the cost of running services and the assumptions about training data. Technology and law move at different speeds. We need to watch both at the same time.