What makes a coding assistant "agentic"

Two ways to get help writing the same line

Type def published? inside Article, and an older autocomplete tool guesses the rest of the line, maybe published_at.present?, right in the editor, from patterns it has seen. It never opens another file, never runs anything and never checks whether the guess was right. You accept the suggestion or you don't, one keystroke at a time.

Now ask a newer kind of tool, in a chat panel instead of inline, to "add a published? method to Article that returns true once published_at is set." It reads the file first and writes the method. If it has permission to run commands, it can open a terminal, run the test suite, see a failing test and edit the method again before showing you anything. That loop of read, write, run, check and repeat, without you typing each step, is what "agentic" means here.

The word names a behavior, not a brand

"Agentic" isn't one product's name. It describes any tool built around the read-write-run-iterate loop, whoever makes it. Lessons 2 and 3 look at two current examples, Cursor and Antigravity, but the loop, not either brand, is what this course teaches.

Why the difference is bigger than it sounds

Older autocomplete only proposes the next few characters where your cursor sits. An agentic tool can decide on its own that fixing a bug means touching three files you didn't mention, and then touch them. That's a jump in capability. It's also a jump in how much a developer has to check before trusting the result, which is the subject of lessons 4 to 7 and of the next course.

What this course covers

1. This lesson: the read-write-run-iterate loop that sets agentic tools apart from autocomplete.
2. Cursor, one current example, described from its documentation.
3. Antigravity, a second current example, described the same way.
4. What these tools can touch: files, terminal commands, and the limits of both.
5. The shift this creates in a developer's job.
6. One concrete request, and how to read the answer critically.
7. A wrap-up, and what comes next.

By lesson 7, "agentic" is a specific, checkable behavior you can recognize in any tool, this year's or a future one.