For years, artificial general intelligence, or AGI, has been presented as a future horizon: the moment when machines would stop excelling only in specific tasks to perform generally in a wide variety of intellectual activities.
Now some of the most relevant voices in the technology industry are beginning to talk about it in the present tense.

“The AGI has arrived”said this weekend Jensen Huang, Founder and CEO of Nvidia, after the launch of GPT-6 Astra, the new OpenAI model. The company run by Sam Altman has not officially made such a categorical statement, but its President and Co-Founder, Greg Brockman, has positioned the launch as a possible turning point.

“Everyone has a different definition of AGI. It’s kind of gray and fuzzy. But I think when we look back, people will think it was around this time and with this model.”Brockman said after the presentation, according to the Financial Times.

The question thus transcends the discussion about whether GPT-6 Astra can formally receive the label of artificial general intelligence. The possibility that is beginning to arise is that this period will end up being considered retrospectively as the beginning of the AGI era.

An AI designed to do, not just respond

The launch that triggered these statements occurred on September 3. OpenAI defines GPT-6 Astra as its model “smarter and more aligned” so far and especially highlights his abilities in the use of computers, programming, cybersecurity, science and professional work.

One of the fundamental differences is precisely in their capacity for action. And Astra can browse the Internet, fill out forms, update records in a CRM, organize calendars, research online information, work on documents, analyze scientific data, generate graphs, create web pages, perform quality controls on them or install and test software autonomously.
You can also tackle professional multi-step workflows and produce finished documents, spreadsheets or presentations. When using computers, OpenAI claims to complete those tasks approximately 47% faster.

Additionally, Astra is the first OpenAI model to reach the “Critical” level in cybersecurity capabilities within its Preparedness Framework. According to the company, with the right tools and access you can locate unknown vulnerabilities and develop new ways to exploit them in well-protected systems without a person having to direct each step. OpenAI claims to have reinforced security measures against these capabilities.

What does it really mean to have achieved AGI?

There is still no universally accepted definition of artificial general intelligence, so there is no definitive test capable of determining the exact moment of its arrival.

One of the definitions used in recent years comes from investor Vinod Khosla: we would reach AGI when artificial intelligence could perform 80% of the work needed in 80% of the economically valuable jobs in the world. Under a criterion of this type, it would still be difficult to maintain that the threshold has been exceeded.

Other approaches place more importance on autonomy, generality, and the ability to use tools.
Veteran technology entrepreneur Bill Nguyen maintains that the transition began months ago and considers a capability that he sees in both Astra and Anthropic’s more advanced models to be especially significant: they no longer need to receive all the tools explicitly, but can instead locate appropriate resources and programs to achieve a goal. Thus, a person proposes what they want to achieve and the system determines what tools they need, interacts with them and executes the work. For Nguyen, developers have brought the models closer to AGI precisely by training them in the use of tools and progressively increasing their autonomy. The limit, he maintains, begins to shift from the capacity of the model to the ambition of people and organizations to deploy it.

In this sense, Jensen Huang’s statement after GPT-6 Astra does not appear in isolation. In Nvidia’s results presentation last August, the executive had already stated that “for many tasks” AGI could be considered to have been achieved. Their argument was that the discussion about labels is becoming less important than looking at what useful and economically valuable work systems are capable of doing. His vision points towards organizations in which people will coexist with much larger numbers of digital agents. Huang has raised as an example the possibility that a company like Nvidia, with tens of thousands of human workers, could operate in the future with hundreds of thousands or even millions of AI agents continuously executing tasks. It is an evolution that shifts the focus from AI that responds to AI that acts.

But not everyone shares Huang’s enthusiasm. Researcher and industry critic Gary Marcus has questioned his statement, pointing out that proclaiming the arrival of AGI without establishing a definition and clear evidence contributes to confusing a scientific debate that remains open.

Even within OpenAI there is caution around the term. The official presentation of GPT-6 Astra itself avoids explicitly defining the model as AGI and speaks of it as a “new generation of intelligence”.

Contradiction helps explain the current moment: capabilities advance faster than the consensus necessary to name them. Perhaps that is why the idea raised by Brockman is more relevant than trying to establish an exact date. There may not be an unequivocal moment at which the industry can declare that there was no AGI yesterday and there was today.

And the signs begin to accumulate: models that use computers, search for tools, conduct research, produce finished professional work, execute multi-step processes, and are given increasing autonomy to decide how to achieve a goal.
But beyond determining whether GPT-6 Astra is officially an AGI, companies and professionals will have to consider what happens if we have already entered a stage in which machines can take on an increasing share of the intellectual and operational work