← Retour au blog
tech 4 September 2026

Which Tools Do Claude, Codex, and Cursor Choose? Insights from 17k Runs

Armature Research examined nearly 17,000 sessions to understand how coding agents choose tools. Discover key findings and their impact on developers and tool vendors.

Article inspired by the original source
Which tools do Claude, Codex and Cursor choose? We measured 17k runs to find out ↗ armature.tech

Introduction

In a world where coding agents are increasingly taking over, understanding how they choose tools is crucial. Armature Research conducted a massive study on 16,893 sessions to uncover how agents like Claude Code, Codex, and Cursor make these choices.

Why It Matters?

Coding agents no longer just recommend solutions; they implement them. For developers, this means they need to trust their agents' choices to integrate the right tools into their projects. For tool vendors, being chosen by a coding agent can spell the difference between success and failure.

The Experiment

Armature tested agents across various contexts: from "vibe-coders" to senior developers, with various prompts and needs. For instance, an amateur user might ask Claude Code how to store data for a personal travel app. Claude recommends Neon for its simplicity and free tier. A senior engineer asking the same question to Cursor receives the same recommendation, proving consistency in tool choices despite different contexts.

The Results

The results show that Neon was the most recommended database tool. Agents also revealed preferences for other tool categories, determined by factors such as ease of integration, predictable costs, and full management.

Implications for Developers

For developers, this means they can rely on agents for relevant and efficient tool recommendations. However, understanding the limits and underlying criteria of these recommendations is crucial.

Implications for Tool Vendors

Vendors must focus on optimizing their products to appeal not only to end-users but also to the agents recommending their use. Transparency, documentation, and easy integration are significant assets.

Conclusion

Armature's study is a first step in understanding how coding agents influence tool choice. For developers and vendors, adapting to this new reality is essential. Let's discuss your project in 15 minutes.

coding agents developer tools automation Neon database AI recommendations
Deepthix newsletter · 100% AI · every Monday 8am

An AI agent reads tech for you.

Our AI agent scans ~200 sources per week and ships the best articles to your inbox Monday 8am. Free. One click to unsubscribe.

Visit the newsletter page →

Want to automate your operations?

Let's talk about your project in 15 minutes.

Book a call