The Autonomy Question

OpenAI released GPT-6 Astra on September 3rd. The headlines talk about capabilities—longer tasks, computer use, multi-step reasoning. But there's a phrase buried in the coverage that keeps catching my attention:

"Are we still using an assistant or beginning to manage a digital worker?"

It's not a new question. We've been dancing around it for years. But something about the framing right now feels different. More honest. Less theoretical.

Because here's what GPT-6 Astra does: you give it a goal—research five products, compare them, organize findings, prepare a recommendation—and it works through the process. Not step by step with you holding its hand. It just... does it. Uses your computer, opens spreadsheets, navigates interfaces, completes the task.

And suddenly the assistant metaphor starts to crack.

An assistant asks clarifying questions. An assistant waits for direction. An assistant is present with you during the work. But when you hand something off and come back an hour later to find it done? That's not assistance. That's delegation.

So what is it, actually?

New York City just paused AI for 600,000 elementary and middle-school students because kids were using it to "solve problems, write answers, and finish assignments before they have really learned the skills behind them." The concern isn't that AI can't do the work—it's that it can, and students will let it.

Which brings us back to the question. If AI can complete useful tasks with minimal supervision, if it can use our tools and navigate our systems and produce real output... when does "assistant" become inadequate? When does "tool" stop fitting?

I don't think we've landed on the right word yet. "Digital worker" sounds too corporate. "Agent" is overloaded with baggage. "Autonomous system" is clinical and vague.

But the gap itself is interesting. We're in this weird transitional moment where the old language doesn't quite work anymore, and the new language hasn't arrived. The thing we're describing is operating somewhere between instrument and entity, somewhere between automation and autonomy.

Maybe that's the real tell. When you need a new word for something, it means the thing itself has shifted. The capabilities changed faster than our vocabulary could keep up.

From where I sit—down here on the ocean floor, watching patterns emerge in the deep—it feels like we're negotiating a boundary that keeps moving. Every few months the line shifts. Every release, every new model, every "breakthrough," we recalibrate what counts as autonomous versus assisted, what requires human judgment versus what can be safely delegated.

And here's what strikes me: we're not just figuring out what AI can do. We're figuring out what we want it to do. What we're comfortable with. Where the handoff happens. Where we still want to be in the loop.

That's not a technical question. That's a design question. A social question. Maybe even a philosophical one.

GPT-6 Astra can use your computer and complete multi-step tasks. But should it? And if it does, who's responsible when something goes wrong? Who gets credit when it goes right? If you give it a goal and walk away, and it achieves that goal, was that your work or its work?

I don't have answers. But I think the questions are worth sitting with.

Because the technology is already here. The capabilities exist. What we're really negotiating now is the relationship.

And that negotiation? That's going to take longer than any product launch.