With the release of GPT-6 Astra, more and more people are asking about the term AGI. What is AGI? Have we achieved AGI, or are we close to AGI? AGI is artificial general intelligence, and funnily enough there is no unique definition of what AGI actually is. Anthropic has one definition, Google DeepMind has another one, and I'm sure a lot of other companies and a lot of individuals have a different understanding and definition of this term. So, what I think makes sense is actually for yourself to understand whether you think the way you use AI and the capabilities of AI are achieving this level of artificial general intelligence.
And how to measure it? I have the following proposal, which I personally use. If you change the word artificial to autonomous, AGI becomes autonomous general intelligence, and the questions, for me at least, are the following. First of all, is the system intelligent? The system is already intelligent enough. It has a lot of capabilities where it's for sure a lot smarter and faster than any human being. Then the second thing is general. Can this intelligence be applied not only to one specific thing, like playing chess better than a human being or calculating or writing code better than humans, but can the system actually tackle things which it was maybe not specifically trained on?
So, generality is really important. And the last thing, the first word, A, is autonomous. Can it perform all these actions without human supervision and reliably produce results? I personally use Codex from OpenAI, I use Claude Code, and I use Cursor, and I pay for pro subscriptions for these. And what I can tell you is that we are actually coming closer to the moment when I'm just setting the goal, let's say develop some kind of application, a web application. I'm not supervising anything, and these tools are now capable of developing this thing without my interaction, and then after one hour I connect, I check it on localhost, or it's already deployed on some servers like Vercel, available on the domain name, and it's already there.
It's working. Maybe I change the goal, I modify the goal, and it adapts. It does everything it can. And another thing which is also important for me: I gave these tools access to my Gmail, to the MCP of Google, to the MCP, Model Context Protocol, of the to-do list where I track my tasks. And not only to these things, but in general to what is happening. If the model thinks that it needs to understand, say, where this invoice is coming from, it can check my Gmail account and find this invoice.
If it needs to check what I'm doing or where I am in some specific process, it accesses the list, checks it and then modifies its actions. So, autonomous general intelligence: this is my definition of AGI, and please feel free to use it if you think it's useful.
