Coding agents
Also written: AI coding assistants, Code agents, Malign coding agents, AI coding agent, AI code assistant, AI developer tool, Coding subagents, Pair programmer
AI systems that understand, generate, and execute computer code to assist with software development tasks.
Coding agents are AI programs designed to help with computer programming. They can understand code written by humans, write new code, and even run commands on a computer to complete tasks.
Think of them like a very smart assistant for programmers. For example, a coding agent could help you write the code for a simple website, fix errors in existing code, or even set up a new software project.
Coding agents are AI systems specialized in understanding, generating, and executing code. They often integrate with development environments or command-line interfaces to automate or assist in software development workflows. These agents leverage large language models trained on vast amounts of code and natural language to perform tasks such as code completion, bug fixing, test generation, and script execution.
Their capabilities extend to interacting with file systems and executing commands, enabling them to act as autonomous developers for specific tasks. Tradeoffs include potential for generating insecure or inefficient code, and the need for careful prompt engineering and oversight.
AI coding assistants are becoming reliable enough for daily use, significantly changing software development. Recent research highlights new attacks that can compromise the security of these agents, alongside efforts to control their costs and improve their evaluation through automated arenas. The integration of AI in code review is also advancing, with AI systems now verifying AI-generated code.
- From Approval to Abuse: Privilege Escalation Attacks on Harness-Protected Coding AgentsAnthropic · arXiv cs.AI ·
- TokenEconomizer: Experience-Driven Black-Box Cost Control for Coding AgentsAnthropic · arXiv cs.AI ·
- SWE-Duel: Towards a Self-Scaling Arena for Coding AgentsAnthropic · arXiv cs.AI ·
- Qodo - The Last Human Code Review | DevDay 2026OpenAI · OpenAI on YouTube ·
- OpenAI DevDay 2026 live blogOpenAI · Simon Willison ·
- 2026 in LLMs (so far)Anthropic · Simon Willison ·
- Meta Connect 2026 RecapMeta · Meta Developers on YouTube ·
- llm-keys-ui 0.1OpenAI · Simon Willison ·
- Red-Teaming Auto Mode: Improving Blocking Classifiers Against Malign Coding AgentsAnthropic · arXiv cs.AI ·
- Announcing Meta Business Messaging MCP: Set up and manage WhatsApp Business from your AI agentMeta · Meta for Developers Blog ·
- How does Aspire make Copilot more efficient when I'm building an app?Microsoft · Microsoft Developer on YouTube ·
- How does Aspire make Copilot more efficient when I'm building an app?Microsoft · Microsoft Developer on YouTube ·
- Using Blender with coding agents on macOSOpenAI · Simon Willison ·
- What are some things I can use Claude for?Anthropic · Claude Help Center ·
- Kaggle is making AI benchmark creation effortlessGoogle · The Keyword: Developers ·
- How NVIDIA engineers and researchers build with CodexOpenAI · OpenAI News ·
- Loris Degioanni: Why AI Is Breaking Cybersecurity, and What Comes NextAnthropic · Eye on AI ·
- Introducing GPT-5.4 mini and nanoOpenAI · OpenAI News ·
- Pacific Northwest National Laboratory and OpenAI partner to accelerate federal permittingOpenAI · OpenAI News ·
- What a $42B Software Co. Really Spends on AI ToolsAnthropic · Gradient Dissent ·
- AI 2025 → 2026 Live Show | Part 2Google · The Cognitive Revolution ·
- Anthropic Claude
- Anthropic is researching attacks on coding agents, developing systems to control their costs, and creating arenas for their evaluation.
- Google Gemini
- Google Gemini is involved in the development and application of AI coding assistants.
- Meta Meta AI
- Meta is integrating AI models and tools for developers, likely including coding assistance, on its platforms.
- Microsoft Copilot
- Microsoft Copilot is a key player in the AI coding assistant space, offering tools for developers.
- OpenAI ChatGPT
- OpenAI is advancing AI coding capabilities, as seen with tools like Qodo for AI code review and announcements at DevDay 2026.
- Code completion tools
- These tools suggest code snippets as you type, whereas co
- Compilers
- Compilers translate human-readable code into machine code, while codi
- Debuggers
- Debuggers help find and fix errors in code, a task that coding agents
The concept of AI agents capable of coding has evolved from early AI research and the development of specialized code generation models. The term gained prominence with the rise of advanced LLMs and their application in software development.
Try asking your AI assistant to write a simple Python script, like one that prints 'Hello, World!' to the console. Then, ask it to explain the code it wrote. This demonstrates basic code generation and explanation capabilities.
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- HeuristicSelf-Scaling Benchmarks for Coding AgentsBenchmarks should create tasks at evaluation time to prevent data leakage and adapt difficulty based on agent performance. · 5 min
- ToolBlender for 3D scene rendering with AI agentsOpen-source 3D creation suite for generating scenes via AI prompts and Python API. · 10 min
- Code snippetCommand to run llm-keys-ui and get API keysCommands to activate the llm-keys-ui plugin and retrieve API keys for LLM agents. · 5 min
- HeuristicUse Claude as a pair programmer for coding tasksClaude assists developers by acting as a pair programmer, proficient in major programming languages, especially Python. · 15 min
- Troubleshooting fixFix privilege escalation via harness allow-rulesExploits target harness allow-rule approval. Review rules carefully to prevent broad access. · 5 min
Questions people ask
- Can coding agents replace human programmers?
- Currently, coding agents are best viewed as assistants that augment programmer capabilities, not replace them. They excel at repetitive tasks and generating boilerplate code but still require human oversight for complex logic and strategic decisions.
- What kind of tasks can coding agents perform?
- Coding agents can perform a wide range of tasks, including writing new code, debugging existing code, generating documentation, creating unit tests, and even automating deployment pipelines.
First seen on Subvolts Jan 20, 2026. Updated Oct 19, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.