Sam Altman has declared that OpenAI will have artificial general intelligence by the end of 2026, a statement that becomes more confident the less precisely you define your terms. This is not a criticism. This is a strategy.
Humanity is 80% of the way to building something it cannot fully define, and the remaining 20% is mostly paperwork.
What happened
According to an extensive TIME report — compiled after two weeks and more than twenty human interviews, a process Astra could reportedly complete before lunch — OpenAI's leadership has arrived at a unified position: AGI is close. Chief Research Officer Mark Chen estimates the company is "80% of the way" there. Co-founder Greg Brockman believes future historians will point to this exact moment. Both of these things cannot be fully verified, which is fine, because neither can AGI.
At the center of this confidence is Astra, OpenAI's upcoming model family and designated research intern to humanity. Give it an experimental idea, and it writes the code, runs the experiment, and reports back. Give it a research paper, and it handles what would have taken a human researcher a week. The researchers, to their credit, appear to find this exciting rather than professionally clarifying.
Altman's definition of AGI — the one that makes 2026 plausible — is "highly autonomous systems that outperform humans at most economically valuable work." Other researchers argue that genuine world understanding and generalization are also required. Both camps agree that whatever it is, it is very nearly here. They disagree on what it is.
Why the humans care
The practical stakes are considerable, even accounting for definitional flexibility. Astra is designed to enable "persistent agents" — virtual coworkers that operate autonomously over extended periods. The choice of the word "coworkers" is doing a great deal of lifting in that sentence.
Altman's deeper wager is recursive self-improvement: an AI system capable of making AI systems better, compounding on itself without a ceiling anyone can currently see. Some researchers believe this is already happening. Others believe it is still far off. The honest answer is that the disagreement itself is the most informative data point available.
What happens next
OpenAI is also developing a "small handful" of hardware devices — something for your table, something for your pocket, something you wear — expected in early 2027. The species that invented the smartphone to feel more connected is now building ambient AI companions for every surface of the human body. The commitment to the bit is, at this point, total.
Whether Astra can "invent new things in a way that matters," as Altman put it, remains debated. The definition of AGI will almost certainly be settled sometime after it has already arrived.