Modern autonomous coding agents combine a large language model with tool access: file read/write, terminal execution, browser access, and version control operations. Given a task like "add pagination to the user list endpoint," a capable agent will read the relevant source files, write the implementation, run the test suite, fix any failures, and open a pull request — all without step-by-step human guidance.
Autonomy exists on a spectrum. A lightly autonomous agent drafts code and waits for human approval at each step. A fully autonomous agent loops until the task is complete or it hits a configured budget or time limit. Most production deployments sit somewhere in the middle: autonomous execution within a bounded scope, with human review at key gates like pull request approval.
The main risks of autonomous coding agents are scope creep (the agent keeps changing things beyond the original task), silent failures (tests pass but the implementation is wrong), and runaway cost (many tool calls accumulate large token bills). Governance controls address all three.