# Code Agent

> **TL;DR.** A code agent is an AI system that executes software engineering tasks autonomously — reading files, editing code, running tests, and iterating toward a goal without per-step human guidance.

- **Category:** AI / Developer Tools / Agents
- **Stage:** validating
- **Age:** 54 days
- **Origin date:** 2026-04-23
- **First detected:** 2026-05-13
- **Canonical URL:** https://earlyterms.com/term/code-agent
- **Sources:** 8 primary URLs

## Definition

A code agent is an AI system that executes software engineering tasks autonomously — reading files, editing code, running tests, and iterating toward a goal without per-step human guidance. Unlike inline code completions, a code agent holds a persistent task context and closes a reason-act-observe loop until the work is done or stalled.

The self-identifier crystallized on [April 23, 2026](https://openai.com/index/introducing-gpt-5-5/) when GPT-5.5 began powering OpenAI's Codex, whose system prompt explicitly frames the model: "I am a code agent (Codex / GPT-5.5) executing real engineering tasks in the user's local repository." That sentence — model naming itself in the first person as an agent class — marks when "code agent" shifted from a category descriptor into a first-person identity claim.

## Example

A developer assigns a GitHub issue to Codex. Codex reads the repo, opens a branch, edits files across three directories, runs the test suite, fixes two regressions, and submits a pull request — all without the developer touching a keyboard. This is the full code-agent loop: goal → plan → act → observe → revise → deliver.

## Analogy

Think of it as a contractor who shows up with their own tools, reads your blueprint, and hands you a finished patch.

## Why it's emerging now

GPT-5.5 launched April 23, 2026 as the engine behind Codex, and its system prompt adopted the first-person declaration 'I am a code agent.' At the same time, security researchers (Pillar, April 2025), paradigm essays (Rijnard, Jan 2026), and 82k-star OSS adoption signal that the term is reaching definitional lock-in across labs, toolchains, and now — model identity itself.

## Related terms

- *alias:* Coding Agents
- *parent:* Agentic Coding
- *related:* Agent Harness
- *related:* AGENTS.md
- *child:* Claude Code
- *child:* OpenAI Codex CLI
- *related:* Agent Loop
- *related:* Dev Agents
- *related:* Managed Agents
- *related:* Cloud Coding Agents
- *related:* Workspace Agent
- *related:* Agent Teams

## Sources

1. [OpenAI: Introducing GPT-5.5 (Codex integration, Apr 2026)](https://openai.com/index/introducing-gpt-5-5/)
2. [TechCrunch: OpenAI launches Codex, an AI coding agent, in ChatGPT (May 2025)](https://techcrunch.com/2025/05/16/openai-launches-codex-an-ai-coding-agent-in-chatgpt/)
3. [GitHub Blog: GitHub Copilot — Meet the new coding agent (May 2025)](https://github.blog/news-insights/product-news/github-copilot-meet-the-new-coding-agent/)
4. [Pillar Security: New Vulnerability — Hackers Can Weaponize Code Agents (Mar 2025)](https://www.pillar.security/blog/new-vulnerability-in-github-copilot-and-cursor-how-hackers-can-weaponize-code-agents)
5. [Rijnard van Tonder: The Code-Only Agent (Jan 2026)](https://rijnard.com/blog/the-code-only-agent)
6. [OpenHands: Coding Agents vs Code Completion (Jan 2026)](https://openhands.dev/blog/agentic-coding-vs-code-completion)
7. [GitHub: openai/codex — Lightweight coding agent, 82k stars](https://github.com/openai/codex)
8. [OpenCode.ai — open source AI coding agent](https://opencode.ai/)

---
_Generated by EarlyTerms · https://earlyterms.com/term/code-agent_