Agentic coding: what it is and how it works

Agentic coding

Agentic coding is a way of building software in which you give an AI agent a whole task rather than single questions. The agent plans the steps, reads and changes the code, runs commands such as tests and fixes what fails until the task is done. You set the goal, decide what the agent may do on its own and review the result.

What agentic coding means

In agentic coding, you hand an AI agent a whole task, such as a bug fix, a new feature or a set of tests. The agent works on it in a loop:

it plans the steps the task needs;
it reads the code it needs and changes files;
it runs commands, such as the tests or the build, and reads their output;
it fixes what failed and repeats until the task is done or it needs a decision from you.

Your part changes with it. You describe the goal and what must be true when the task is finished, decide what the agent may do on its own and what it must ask about, and review the result before the work counts as done. A task written down in advance, with criteria the result is checked against, is the heart of spec-driven development.

Agentic coding, coding agent and AI agent

The three terms name different things:

an AI agent is any AI program that pursues a goal in several steps and uses tools for it, in code or in any other field;
a coding agent is an AI agent made for code: the tool that reads and changes files and runs commands in your project;
agentic coding is the way of working in which you hand development tasks to such an agent and keep the decisions about the goal and the result.

Agentic coding in featureloom

featureloom is a plan board you connect your own Claude Code, Codex or Cursor to. Here the agent creates epics, writes plans and takes them in order, and you decide what it may do.

The order comes from the board: the agent takes the next ready plan, and the server refuses to start a plan that still waits on another. When the agent reports a plan delivered, a person decides on acceptance, not the agent: they accept the work or send it back, and after a rejection the agent goes back to rework. The How featureloom works page shows a plan’s whole road.

Questions

Does agentic coding mean the agent works without me?

No. You set the task, decide what the agent may do on its own and judge the result. In featureloom the agent takes plans and delivers them by itself, and you accept the work or send it back for rework.

How is agentic coding different from code completion?

With code completion, you write the code and the AI suggests how it goes on as you type. In agentic coding, you describe the task and the agent takes the steps: it changes several files, runs the tests and corrects its own work, and you review the result.

Where does the agent’s task come from in featureloom?

From a plan on the board. The app generates a plan from a pasted note or conversation or from a PDF or Word file, and an agent can also propose one from a conversation in its console. For a named plan, the agent receives the role prompt, the whole plan and the documents of the delivered plans it directly depends on, in one call.