GPT-6 Astra Is Here. So, What Will We Do?

GPT-6 Astra is officially here. It can navigate browsers, click buttons, use different software and complete multi-step tasks without needing a human to approve every single action.

My first thought was basically: this is impressive. My second was: what does this actually mean for people at work? I’ve been thinking about this quite a bit since graduating.

One thing that stands out to me is that AI is moving further away from just being a tool we use and closer to becoming something that can actually participate in the workflow. And I think that distinction matters.

If you look at most office jobs, there are probably a lot of tasks that aren’t particularly difficult, but still take time. Updating systems. Filling out forms. Collecting information. Running standard analyses. Moving information between different platforms.

These things are necessary, but they aren’t necessarily where we create the most value. So if an AI agent can take care of them, what happens?

Hopefully, we get more time to think. More time for strategy, creativity, problem solving and actually talking to people. At least that’s the optimistic version. But I think there is another change happening too.

AI agents can potentially run workflows for much longer than a normal interaction with a chatbot. You could start an audit, data analysis or technical process and let it continue in the background. Overnight, even.

That means the traditional workday could start to look different. Instead of opening your computer and starting the day’s tasks, you might open it and see what your AI agents completed while you were away. And then your job becomes reviewing, questioning, changing direction or deciding what is good enough.

If AI becomes increasingly good at executing instructions, then simply knowing how to perform a task becomes less of a differentiator. Knowing what to ask for becomes more important. But even more importantly, knowing whether the result is actually good. That requires context. Judgement. Some level of intuition. Understanding people. Understanding consequences. Ethics. Things that are difficult to reduce to a simple instruction.

I don’t think this means technical skills become irrelevant. Quite the opposite. You need enough technical understanding to know what the AI is doing and when something has gone wrong. But maybe the valuable combination is becoming more about technical understanding and judgement.

There is also the less exciting side of all this. The more AI can actually do, the more risk we introduce.

If an AI can move between systems, access information and make decisions without constant human approval, organisations need to be much more intentional about security and governance. Permissions, monitoring, human oversight and clear boundaries become essential.

So I’m optimistic about agentic AI, but not blindly optimistic. I don’t think the future is humans sitting around while AI does everything. I think it is more likely that the definition of “doing the work” changes. And honestly, as someone just entering the workforce, that is both exciting and a little unsettling. We may be moving from a world where being good at using software is a major advantage to one where being good at directing, questioning and making decisions around software is the real advantage.