What is agentic coding?

Agentic coding is software development where an AI agent plans and runs multi-step tasks on its own, editing files, running commands, and iterating, rather than only suggesting completions. The developer supervises and reviews instead of writing each line.

Agentic coding vs autocomplete vs chat

These sit on a scale of autonomy.

• Autocomplete suggests the next token or line while a person types. The developer drives.
• Chat answers questions and returns snippets the developer pastes in. A person still makes the edits.
• Agentic coding takes a goal. The agent plans steps, edits files, runs commands, reads results, and iterates until the goal is met.

How a coding agent works

An agent runs a plan-act-observe loop. It gathers context from the repository, forms a plan, makes a change, runs something to check it, reads the result, and adjusts.

• Tools: file edits, shell commands, test runs, and search.
• Context: the files, conventions, and history it can read.
• The loop: act, observe the outcome, correct, repeat.

Where it helps

Agents are strong on work that is large but well defined: multi-file refactors, framework migrations, and repetitive changes across a codebase. The breadth that makes them useful is also what raises the review load when the output is not checked.

The review and governance gap

An agent produces more code per hour than a person, but review capacity stays flat. The space between code written and code reviewed is where defects and AI slop collect. Closing it takes AI code review on every change and a layer of code governance that defines what is allowed to merge.