What is Coding agent?

A coding agent is an AI agent tuned for the software-development loop. Given a task, fix this bug, add this feature, refactor this module, it explores the repository, proposes and applies edits, runs builds and tests, reads the output, and iterates until the change works. Tools like Claude Code, Cursor's agent, and similar IDE and CLI agents fall in this category. What makes them effective is the breadth of context and tools they can reach: the file system, a terminal, version control, and increasingly external systems through MCP servers, an issue tracker to read the ticket, a documentation server to look up current API usage, a code host to open the pull request. Because models do not always know the latest library APIs, a coding agent paired with an up-to-date docs server can ground its edits in current references instead of stale training data. Memory raises the ceiling further: a coding agent with a shared memory layer remembers project conventions, past decisions, and gotchas across sessions, so it stops rediscovering the same constraints, which is precisely the gap Glen's org-shared memory is built to close.