AGI capabilities: unlimited domains a general mind can take on.

A capability list is evidence, not a boundary. Unlimited domains show that one intelligence holds across very different kinds of work; they do not mean the twenty-first is impossible. Skills and MCP connections extend the reach, which is why the honest framing is examples of a general intelligence rather than a fixed feature matrix.

What makes a domain an AGI capability rather than a narrow feature is the loop around it. The intelligence takes a goal, plans the route, acts in real systems, verifies the result and reports. The same loop runs whether the work is a website, a server or a report, and that sameness is the point. The row of domains is the surface; the loop is the claim.

Read the list with the six tests in mind: generality, autonomy, real-world operation, self-verification, memory and orchestration. Any entry that cannot be checked against those tests is worth less than one that can.

The list matters most where entries touch. A market report links research to data to documents to content. A site launch runs from research through copy, design, code, CMS and a live check. A maintenance contract runs browser checks, SEO audits, code fixes and reports on a schedule.

Because the intelligence is one system, the output of one domain becomes the input of the next without a human copying it across. That is what removes the integration tax: the re-explaining, the reformatting and the hand-offs that a stack of narrow products pushes onto the person using them. Most of the value sits in those joins, not in the list.

It also means a job can start in one domain and end in another, which is how real work arrives. You do not usually need a report and then, separately, a page. You need the report to become the page, and the page to become a live site with the evidence attached.

Each domain on the list is checked against the same six tests. Generality asks whether one intelligence holds across the domains rather than a separate product per task. Autonomy asks whether a goal becomes a verified result without step-by-step prompting. Real-world operation asks whether something outside the interface changed: a file, a deployment, a live page.

Self-verification asks whether the system checked its own output and reported what failed. Memory asks whether the second job starts from what the first learned. Orchestration asks whether a large goal can be split across agents, run in parallel and merged into one accountable report.

A domain that cannot answer those questions is a feature with a marketing name. The point of listing twenty is not to impress with the count; it is to give a reader enough surface area to test the claim. Pick any two adjacent domains and ask for one job that crosses them. The answer is the AGI capabilities claim, demonstrated or not. It is also the quickest way to compare two products claiming the same list: ask each to show the join between two domains, not the list itself.