Autonomous Agents: Can They Ship Alone?

The frontier: agents that don't just autocomplete but plan, code, test, and open the pull request themselves.

The top 3

  1. Devin: the first 'AI software engineer': Cognition's Devin (2024) promised end-to-end autonomy; early independent runs solved well under 20% of real tasks.
  2. Background & async agents go mainstream: Claude Code, OpenAI's Codex, and Cursor's background agents now run tasks asynchronously and open pull requests unattended.
  3. The trust gap: review, security, and slop: Autonomy stalls on human review — hallucinated APIs and insecure generated code keep engineers firmly in the loop.

Sources

Open the full topic