New AI Coding Agent Builds Itself in 15 Minutes
The XDACode project demonstrates how autonomous agents can now architect their own software systems and integrate OAuth protocols.
Updated on Oct. 21, 2026 in Artificial Intelligence
Live Poll
Do you trust AI agents to independently manage and build software projects for you?
Developers have successfully used the Claude Code coding agent to build a functional, terminal-based AI agent named XDACode in just 15 minutes. This project demonstrates how AI agents can now generate their own complex software architectures and integrate new authentication protocols, such as ChatGPT Pro via OAuth, with minimal human intervention.
Why it matters
This development marks a shift where AI agents are capable of bootstrapping their own successors, potentially accelerating the development cycles for autonomous programming tools. It underscores a growing trend in software engineering where the agent-to-agent feedback loop replaces manual boilerplate coding.
The initial build of XDACode was completed in 15 minutes, with subsequent ChatGPT Pro OAuth integration taking 10 minutes and 59 seconds. The project currently maintains a suite of 54 automated tests, up from 40 in its initial version.
The players
Boris Cherny
A software engineer who has transitioned to utilizing AI agents for 100% of his code contributions.
Claude Code
An autonomous coding agent platform that uses AI models to manage software development workflows.
The details
XDACode is a terminal-based agent written in Python 3, leveraging the Anthropic SDK, Rich for terminal formatting, and prompt_toolkit for interactive interfaces. It includes eight integrated tools for file system manipulation and command execution. A built-in permissions system ensures that all proposed code changes are vetted, while internal mechanisms for context compaction manage the agent's memory usage during extended programming tasks.
Timeline
December 2025: Boris Cherny reports total Claude Code usage for contributions.
June 2026: Boris Cherny stops writing manual code entirely.
October 2026: The project and implementation details are published.
The Tech Race
This project builds directly upon the capabilities of Claude Code to prove that autonomous agents can iterate on their own functional requirements. It follows the precedent set by automated coding environments that mark a departure from traditional manual IDE-based development workflows.
Software engineers and developers can now utilize agentic workflows to handle boilerplate architecture, potentially reducing the time required to build CLI tools from hours to minutes. Users should note that these agents require a permissions-based review process to approve automated file operations.
The takeaway
The ability for an AI to architect and test a new coding agent in under 30 minutes signals that the role of the developer is increasingly becoming one of orchestration rather than direct implementation. Watch for updates on how these self-iterating agents handle complex dependency trees compared to traditional human-written codebases.
Further reading
Explore the broader evolution of automated development tools in our Artificial Intelligence section.
Source note: This article includes information reported by XDA-Developers.
Live Poll
Do you trust AI agents to independently manage and build software projects for you?
