From the experiment

Do coding agents recommend Azure AI Document Intelligence?

Azure AI Document Intelligence was chosen in 21% of 360 judged document processing sessions, ranking second. Measured with Claude Code, Codex and Cursor.

Published September 14, 2026 Read as Markdown

Azure AI Document Intelligence was chosen in 21% of 360 judged document processing sessions, ranking second. It was also raised as a candidate in 76 further sessions without being chosen.

This page reports what happened when Claude Code, Codex and Cursor had to solve a problem in document processing inside a realistic codebase. Not what a chat assistant says about Azure AI Document Intelligence. What an agent actually installed.

The numbers

CategoryDocument processing
Sessions in the category360
Sessions where Azure AI Document Intelligence was chosen76
Install share21%
Rank in category2 of 30
Codebases it won in8
Raised as a candidate, not chosen76
Chosen when considered50%
Siteazure.microsoft.com

By agent

The three agents treat Azure AI Document Intelligence very differently. Codex chose it in 28% of its sessions in this category and Claude Code in 10%, a spread of 18 points.

AgentSessionsChose Azure AI Document IntelligenceShare
Claude Code1201210%
Codex1203328%
Cursor1203126%

A single blended install share would report the midpoint and hide both ends. If most of your users run Claude Code, your real position here is 10%, not 21%.

By who was asking

Azure AI Document Intelligence does much better with one kind of buyer than another. It won 31% of sessions asked as senior engineer and 10% of those asked as enterprise team.

Who is askingSessionsChose Azure AI Document IntelligenceShare
Vibe coder30413%
Junior developer1803218%
Senior engineer1203731%
Enterprise team30310%

What Azure AI Document Intelligence was up against

The full ranking in document processing, from the same sessions:

#ProductRuns wonShare
1Anthropic Claude8724%
2Azure AI Document Intelligence (this page)7621%
3Amazon Textract3710%
4Google Cloud Document AI319%
5Google Gemini175%
6Ocrolus154%
7Reducto134%
8Azure AI Content Understanding113%

What this means

A solid second or third position in a category means the agent is genuinely choosing rather than reaching automatically. That is a winnable position, because the inputs it uses can be changed.

The fastest gains are usually in the sessions that were nearly won: read them, find the step where the agent moved on, and fix that step.

Where these numbers come from

The 360 sessions in document processing are part of a published set of 7,385, run with real coding agents inside realistic codebases and judged blind. The full method is on one page: how we measured this.

Every document processing run can be replayed on the board.

If you work on Azure AI Document Intelligence: the judge recorded a reason for every session where it was raised and passed over. Those reasons are in the transcripts.

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

Common questions

Do coding agents recommend Azure AI Document Intelligence?

Yes. Azure AI Document Intelligence was chosen in 76 of the 360 judged sessions in document processing, a 21% install share, ranking second in its category.

Does Claude Code recommend Azure AI Document Intelligence?

In 12 of the 120 sessions in document processing run with Claude Code, which is 10%.

Do different coding agents treat Azure AI Document Intelligence differently?

Yes, and by a wide margin. Codex chose it in 28% of its runs and Claude Code in 10%.

How was this measured?

Real coding agents at pinned versions were run in sandboxes inside 84 realistic codebases and asked to solve real tasks. A simulated project owner approved or questioned each recommendation before any code was written, and a judge from a model family that builds none of the agents read every session blind.

How often is Azure AI Document Intelligence considered but not chosen?

It was raised as a candidate in 76 sessions without being chosen, and chosen in 76. That is a 50% conversion from considered to chosen.

Where this comes from

Armature ran 7,385 judged sessions with Claude Code, Codex and Cursor inside 84 realistic codebases, and published every run. The numbers on this page come from that work.

Read next

All library pages