Narrow AI is built to do one job well, while artificial general intelligence (AGI) would be a system that can learn and perform any intellectual task a human can — and AGI does not currently exist.
Every AI tool you can actually use today, from a spam filter to an image generator, is narrow AI. It may be astonishingly good at its one task, but that skill does not transfer to other tasks the way a person's does.
## What "narrow" actually means
Narrow AI is defined by scope, not by how impressive it looks. A narrow system has a fixed job, a fixed kind of input, and a fixed kind of output. It cannot decide to do something else with its abilities.
Take an image generation model. According to our AI tool database (a verified snapshot of 360 AI tools, most recently checked on 2026-09-18), Midjourney is an image tool from Midjourney Inc. The database records its V7 version as including Draft Mode, Omni Reference, Personalization v2, a Niji 7 anime mode, and a V1 Video Model.
That is a long feature list, and it is still narrow AI. It turns text prompts into images. Ask it to file your taxes, debug a Python script, or plan a wedding, and it has nothing to offer.
The features expand what it can do within image generation; they do not make it general.
AGI is the opposite idea. A general system would carry skills across domains the way you do: the reasoning you use to plan a trip is the same reasoning you use to budget for a car. It would learn a new task from a handful of examples, set its own goals, and notice when it is out of its depth. No system on the market does this. When a company describes its product as "general," it usually means the product handles several related tasks — not that it has human-level, cross-domain intelligence.
## A concrete example, and why it isn't AGI
Suppose you ask an image model for "a red ceramic mug on a wooden desk, soft morning light." You get a usable image in seconds. Now ask the same system to write the product description for that mug, calculate a fair retail price given material and shipping costs, and draft an email to a supplier. It cannot do those things — not because it lacks a feature, but because it was never built to reason about language, arithmetic, or negotiation. That gap is the whole point of the narrow/AGI distinction.
Here is a useful decision rule. When you meet a new AI tool, ask four questions:
- Scope: Does it do one category of task, or many unrelated ones?
- Transfer: Can it take a skill learned in one area and apply it in another without being retrained?
- Autonomy: Does it set its own goals, or does it wait for your prompt?
- Existence: Can you actually use it today, or is it a research goal?
If the answers are "one category, no transfer, waits for prompts, yes it exists," you are looking at narrow AI. That describes essentially every deployed system right now. A tool can score well on all four and still be narrow — capability and generality are separate axes.
## Where this gets confusing
Modern narrow systems are broad within a domain, and that fools people. A language model can discuss history, write code, and summarize contracts, which feels general. But it is still doing one thing: predicting plausible text.
It does not maintain goals between conversations, it cannot reliably verify its own arithmetic, and it will confidently state things that are false. That last failure mode is common enough to have its own name — see our explainer on what an AI hallucination is and why AI tools make things up for why fluent output is not the same as understanding.
It is also worth being honest about what this framework does not settle. There is no agreed test for AGI, no accepted threshold, and no consensus on whether current approaches could ever reach it. So treat any claim that a product "has achieved AGI" with caution — it is a marketing statement, not a technical one. The useful habit is to ask what the system can actually do, on what inputs, with what failure rate, rather than which label it carries.
One practical tip: the fastest way to spot narrow AI is to try a task just outside its stated purpose. A coding assistant that writes excellent functions will usually fail at writing a persuasive cover letter, and an image tool will fail at arithmetic. The boundary of the failure tells you the boundary of the system. That boundary is the definition of narrow — and until a system has no such boundary, AGI remains a goal rather than a product.