what sign up
nixm nixm

← learn

What is AGI? Artificial general intelligence, explained plainly

October 14, 2026

AGI stands for artificial general intelligence: an AI that can learn and perform most intellectual tasks a human can, across domains, rather than being good at one thing. Today's models are broad but uneven: they can pass professional exams and fail at tasks a child manages. There is no agreed test for when AGI has arrived, which is why serious people disagree about whether it is years away, decades away, or already here in a weak form. (If you were looking for AGI on a tax form, that is adjusted gross income, a different thing entirely.)

What the letters mean

Artificial general intelligence. The important word is general. Every AI before the 2020s was narrow: superhuman at chess and useless at checkers, perfect at reading cheques and blind to a cat. AGI is the idea of a single system with the breadth of a person: it can learn a new domain, reason about something it was never trained on, and move between tasks the way people do.

Two other terms sit around it. Narrow AI is everything we have had for decades. Superintelligence is the hypothetical step beyond AGI, where a system exceeds the best humans at essentially everything. AGI is the middle: roughly human, roughly everywhere.

Why nobody agrees on the test

There is no finish line that everyone accepts. Some definitions are about capability: OpenAI's charter describes AGI as "highly autonomous systems that outperform humans at most economically valuable work". Some are about method: can it learn a genuinely new skill from a few examples the way a person does. Some are economic: when it can do a large share of jobs at human quality and cost. Each definition gives a different date, which is why one expert says 2027 and another says never in the same week, and both are answering a different question.

How close today's models are

Closer than almost anyone predicted in 2019, and further than the demos suggest. Current models are startlingly broad: the same system writes code, passes medical exams, translates, explains physics and drafts contracts. That breadth is new and is the reason the AGI conversation moved from science fiction to boardrooms. But the breadth is uneven. The same model that out-scores most lawyers on an exam can fail to count the letters in a word, cannot learn from one example the way a child does, forgets the start of a long task, and has no goals of its own. The AI vs human intelligence page lays out that jagged shape in detail. Whether you call today's systems "early AGI" or "very good narrow AI with wide coverage" depends on which definition you picked above.

What the people building it say

The major labs state AGI as a goal, in their founding documents, and disagree in public about the timeline: leaders at the frontier labs have given estimates ranging from a few years to the next decade, while many academic researchers argue that current methods will plateau before general intelligence. Both camps include people who built the systems. The honest summary is that the people closest to the work believe it is possible and are not sure when, and that their incentives, commercial and reputational, point in different directions.

Why it matters even if it is not here

Because policy, investment and your own planning run on expectations. The scale of spending on chips and data centres, the pace of regulation, the question of which jobs change: all of it is shaped by how close people think AGI is. And the practical questions do not wait for the definition. An AI that is "only" very broad already changes work, education and privacy today. The is AI dangerous page ranks those near-term risks ahead of the speculative ones for that reason.

What to do with this

Treat AGI as a direction, not a date. The capabilities arriving on the way to it are the thing to use and to be careful with: check what the model says, keep the judgement and the responsibility, and do not hand your private life to a system that keeps it. That last habit holds whatever the timeline; it is why nixm keeps no transcript and trains on nothing you say.

frequently asked

What does AGI stand for?

Artificial general intelligence: an AI that can perform most intellectual tasks a human can, across domains, rather than being specialised for one.

What is the difference between AI and AGI?

AI is any system doing something that needs intelligence, usually in one domain. AGI is the hypothetical system with the general breadth of a person: it can learn new domains and move between tasks the way people do.

Has AGI been achieved?

There is no agreed test, so there is no agreed answer. Current models are very broad but uneven, strong on exams and weak on tasks a child manages. Some call that early AGI; most call it advanced narrow AI with wide coverage.

When will AGI happen?

Estimates from people building it range from a few years to the next decade; many researchers think current methods will plateau first. The dates differ mostly because the definitions differ.

What is the difference between AGI and superintelligence?

AGI is roughly human-level across most tasks. Superintelligence is the step beyond, exceeding the best humans at essentially everything. Neither exists today.

Does AGI mean adjusted gross income?

On a tax form, yes: AGI there is adjusted gross income, your income after certain deductions. This page is about artificial general intelligence, an unrelated use of the same letters.

try it now — no account, no transcript kept. even the signal dissolves in seven days.

seed a signal →

Enter your email and receive a link to sign in. No passwords, just a secure signal to connect.

why connect

  • 20 free credits to start
  • your signals, remembered
  • buy more credits anytime
  • image & video generation features
or
+