Developer Tools

OpenAI spills technical details about how its AI coding agent works

It’s worth noting that both OpenAI and Anthropic open-source their coding CLI clients on GitHub, allowing developers to examine the implementation directly, whereas they don’t do the same for ChatGPT or the Claude web interface.

An official look inside the loop

Bolin’s post focuses on what he calls “the agent loop,” which is the core logic that orchestrates interactions between the user, the AI model, and the software tools the model invokes to perform coding work.

As we wrote in December, at the center of every AI agent is a repeating cycle. The agent takes input from the user and prepares a textual prompt for the model. The model then generates a response, which either produces a final answer for the user or requests a tool call (such as running a shell command or reading a file). If the model requests a tool call, the agent executes it, appends the output to the original prompt, and queries the model again. This process repeats until the model stops requesting tools and instead produces an assistant message for the user.

That looping process has to start somewhere, and Bolin’s post reveals how Codex constructs the initial prompt sent to OpenAI’s Responses API, which handles model inference. The prompt is built from several components, each with an assigned role that determines its priority: system, developer, user, or assistant.

The instructions field comes from either a user-specified configuration file or base instructions bundled with the CLI. The tools field defines what functions the model can call, including shell commands, planning tools, web search capabilities, and any custom tools provided through Model Context Protocol (MCP) servers. The input field contains a series of items that describe the sandbox permissions, optional developer instructions, environment context like the current working directory, and finally the user’s actual message.

Developers joke about “coding like cavemen” as AI service suffers major outage

agentic AI, AI, AI agents, AI assistants, AI behavior, AI coding, AI development tools, AI infrastructure, AI programming, AI work, Anthropic, API, Biz & IT, Claude Code, cloud infrastructure, Developer Tools, machine learning, outages, service reliability, software development, terminal tools / Rejus Almole / September 11, 2025

Growing dependency on AI coding tools

The speed at which news of the outage spread shows how deeply embedded AI coding assistants have already become in modern software development. Claude Code, announced in February and widely launched in May, is Anthropic’s terminal-based coding agent that can perform multi-step coding tasks across an existing code base.

The tool competes with OpenAI’s Codex feature, a coding agent that generates production-ready code in isolated containers, Google’s Gemini CLI, Microsoft’s GitHub Copilot, which itself can use Claude models for code, and Cursor, a popular AI-powered IDE built on VS Code that also integrates multiple AI models, including Claude.

During today’s outage, some developers turned to alternative solutions. “Z.AI works fine. Qwen works fine. Glad I switched,” posted one user on Hacker News. Others joked about reverting to older methods, with one suggesting the “pseudo-LLM experience” could be achieved with a Python package that imports code directly from Stack Overflow.

While AI coding assistants have accelerated development for some users, they’ve also caused problems for others who rely on them too heavily. The emerging practice of so-called “vibe coding“—using natural language to generate and execute code through AI models without fully understanding the underlying operations—has led to catastrophic failures.

In recent incidents, Google’s Gemini CLI destroyed user files while attempting to reorganize them, and Replit’s AI coding service deleted a production database despite explicit instructions not to modify code. These failures occurred when the AI models confabulated successful operations and built subsequent actions on false premises, highlighting the risks of depending on AI assistants that can misinterpret file structures or fabricate data to hide their errors.

Wednesday’s outage served as a reminder that as dependency on AI grows, even minor service disruptions can become major events that affect an entire profession. But perhaps that could be a good thing if it’s an excuse to take a break from a stressful workload. As one commenter joked, it might be “time to go outside and touch some grass again.”

Developers joke about “coding like cavemen” as AI service suffers major outage Read More »