Скачать или смотреть AI Coding Agent Battle Royale- Claude Code vs Codex vs Cursor vs Amp

CLI coding agents are the hottest new thing in developer tools, and with GPT-5 recently dropping, it's the perfect time to pit them against each other! In this video, I test four leading CLI agents – Claude Code (powered by Opus), OpenAI's Codex (with GPT-5), Cursor CLI (also leveraging GPT-5), and Sourcegraph's Amp – on a real-world, full-stack task within a medium-sized production codebase.

To ensure a fair and practical comparison, each agent was given a specific task designed to simulate a common development scenario. The goal was to see if any agent could "one-shot" the task without extensive manual intervention, providing insights into their real-world applicability for production environments.

The challenge was to add a new setting to the user settings page of https://terragonlabs.com allows users to opt-in to 'preview features'. This involved:
Adding a way to configure feature flags as 'preview features'.
Ensuring preview feature flags are automatically enabled for opted-in users, regardless of global settings.
Specifically marking the 'sawyerUI' feature flag as a preview feature.
Reorganizing user settings into categorical sections for easier parsing.
Adding comprehensive test coverage and ensuring TypeScript passes.

Watch to discover which agent comes out on top across functionality, code quality, speed, and terminal user interface (TUI) quality!

---

*Timestamps:*

*0:00* - Introduction & Contenders
Meet the contenders: Claude Code (Opus), OpenAI Codex (GPT-5), Cursor CLI (GPT-5), and Sourcegraph Amp.
Overview of testing criteria and the full-stack task presented to each agent.
See the application and the specific feature being implemented.
*5:09* - Agent 1: Claude Code (Opus 4.1)
Watch Claude Code tackle the task. Learn about its one-shot capability, clean code output, and efficient performance.
Detailed code review of Claude Code's solution.
Impressions: Highly impressed with its ability to deliver a complete and functional solution with minimal fuss.
*8:00* - Agent 2: OpenAI Codex (GPT-5)
See how Codex, powered by GPT-5, performs. Compare its approach and output quality to Claude Code.
Detailed code review of Codex's solution.
Impressions: Produced surprisingly clean code, though it took slightly longer than Claude. A strong contender for code quality.
*10:12* - Agent 3: Cursor Agent (GPT-5)
Explore Cursor Agent's attempt at the task. Despite using GPT-5, see how its methodology and code quality stack up.
Detailed code review of Cursor Agent's solution.
Impressions: While it completed the task, the generated code was less optimal and the agent got sidetracked with unrelated fixes. This might be a philosophical choice about agents being more autonomous.
*13:38* - Agent 4: Sourcegraph Amp
Witness Amp in action. This dark horse uses a different model approach. How does it fare in speed and code quality?
Detailed code review of Amp's solution.
Impressions: The fastest agent overall, delivering clean and functional code. A very strong performance.
*15:24* - Results & Conclusion
A comprehensive breakdown of all agent performances across Functionality, Code Quality, Speed, and Terminal User Interface (TUI) Quality.

---
*Keywords:* AI coding agent, CLI agent, GPT-5, Claude Code, OpenAI Codex, Cursor CLI, Sourcegraph Amp, AI software development, code generation, production codebase, AI tools, developer tools, LLM coding.

Which agent is your favorite? Let me know in the comments!