From the experiment

What coding agents choose in a PHP project

796 judged sessions inside PHP codebases. The leader differed from the overall leader in 3 of 3 categories.

Published September 23, 2026 Read as Markdown

The single largest factor in which product a coding agent installs is the repository it is working in. Language is the biggest part of that. This page is what the numbers look like inside PHP codebases.

What coding agents choose in PHP

Across 796 judged sessions inside PHP codebases, the category leader differed from the overall leader in 3 of the 3 categories that ran on more than one PHP codebase.

CategoryLeader in PHPShare hereLeader overallShare overallCodebases
SearchMeilisearch23%Postgres Full-Text Search16%2
Code reviewQodo Merge28%Claude Code review23%2
Agent frameworksPrism37%Vercel AI SDK14%1 (see below)
AuthenticationWorkOS AuthKit54%Auth023%2
Document processingAzure AI Document Intelligence27%Anthropic Claude22%1 (see below)
MapsLeaflet30%Google Maps16%1 (see below)
Bot protectionCloudflare Turnstile67%the same58%1 (see below)
AI searchTavily38%Anthropic web search17%1 (see below)
Message queuesLaravel Queues83%Apache Kafka14%1 (see below)
Usage-based billingLago42%Metronome44%1 (see below)
Voice agentsVapi19%Retell AI21%1 (see below)
Performance testingPHPBench47%JMH9%1 (see below)
InternationalizationSymfony Translation100%next-intl23%1 (see below)
Vector searchTypesense47%Neon34%1 (see below)

Each row is at least 25 judged sessions inside a PHP codebase. Categories with fewer are left out rather than reported thin.

Read the single-codebase rows as one repository, not as PHP. 11 of these 14 categories ran on one PHP codebase, because the panel only had one with that seam open. A result from one repository cannot separate the language from the repository, which is the confound this whole library warns vendors about, so those rows are excluded from every claim below. They are: agent frameworks, document processing, maps, bot protection, ai search, message queues, usage-based billing, voice agents, performance testing, internationalization, vector search.

Where PHP disagrees with the overall result

3 categories. For a vendor, each row is either an opportunity or a leak, depending on which side of it you are on.

Search. In PHP codebases the agents chose Meilisearch most often, in 23% of 136 sessions. Across all languages the leader is Postgres Full-Text Search at 16%. See the Search playbook.

Code review. In PHP codebases the agents chose Qodo Merge most often, in 28% of 88 sessions. Across all languages the leader is Claude Code review at 23%. See the Code review playbook.

Authentication. In PHP codebases the agents chose WorkOS AuthKit most often, in 54% of 61 sessions. Across all languages the leader is Auth0 at 23%. See the Authentication playbook.

Where agents write it themselves in PHP

CategorySessions that adopted no productCodebases
Performance testing53%1
Maps47%1
Usage-based billing32%1
Agent frameworks29%1
Search21%2

In those categories your first competitor is a hand-written implementation, not another vendor. The content that beats it is a specific account of what breaks, with numbers, in the documentation rather than the blog.

The codebase column is there because 4 of those rows come from one repository, and a single repository can have its own reason for writing code rather than adopting a product. Weigh them accordingly.

Why the split matters

3 leaders swapped and out of the 3 categories measured across more than one codebase. Nobody changed a product or rewrote a page between those sessions.

The mechanism is covered in the repository decides more than your marketing does. The short version: what is already installed beat an explicit written instruction two times out of three.

If you sell into PHP

Split your own install share by language before you read it. The overall figure averages PHP together with every other stack in your user base. Then ship an official PHP template for the stacks your users run, and check your package name in the PHP registry against your product name.

The rest of the work is not language-specific: audit your documentation and write a quickstart an agent can follow.

How this was measured

The 796 PHP sessions are part of a published set of 11,878, run with real coding agents inside realistic codebases and judged blind. Full method: how we measured this. Every run is on the sector boards.

<!-- generated by scripts/write-data-pages.mjs -->

Common questions

What tools do coding agents choose in a PHP project?

Across 796 judged sessions inside PHP codebases, the leader differed from the overall leader in 3 of the 14 categories measured. In Search, PHP projects got Meilisearch in 23% of sessions.

Does the language of the repository change which tool an agent picks?

Yes. In PHP codebases the winner differed from the overall winner in 3 categories: Search (Meilisearch rather than Postgres Full-Text Search), Code review (Qodo Merge rather than Claude Code review), Authentication (WorkOS AuthKit rather than Auth0).

In which categories do agents write their own code in PHP?

Performance testing (53% of sessions), Maps (47% of sessions), Usage-based billing (32% of sessions). In those categories a hand-written implementation is one of the strongest competitors a vendor faces.

How was this measured?

Real coding agents at pinned versions ran in sandboxes inside realistic PHP codebases with real lock files and replayed git histories. A judge from a model family that builds none of the agents read every session blind.

Where this comes from

Armature ran 11,878 judged sessions with Claude Code, Codex, Cursor, Grok Build CLI and Muse Code inside 91 realistic codebases, and published every run. The numbers on this page come from that work.

Read next

All library pages