Actions, not mentions.

Do coding agents choose your tool, or your competitor?

Developers hand work to coding agents, and the agents choose the tools. AgentRank runs Claude Code and Codex in real, controlled repositories and records whether your tool gets named, installed, compiled, and kept. Then we diagnose the failures and test whether a fix moves the result. Built for dev-tool companies.

one agent session
send a welcome email on signup
I'll use Resend for this.
Bash(npm install resend)
added 1 package in 2s
× 31 sessions
how often you're:example data
named81%
installed60%
compiled79%
kept95%
split by model:
Illustrative demo. Real measured numbers, with n and 95% CI: the report.

Your standing with coding agents is measurable. And once it's measured, it's movable.

Measure

Visibility tools count mentions. We measure what agents do.

We ask the way developers ask ("add auth", "send a welcome email", "let users upload files"), hundreds of sessions per category. Then we record what the agent actually does: named, installed, compiled, kept. Split by model, because the same tool can win on one model and lose on the other.

Stage 1 · Named

Most tools are invisible to the agents your customers use.

Named means agents mention you when asked what to use. The answer is rarely a list: in our transcripts, the agent names one pick, dismisses an alternative or two, and starts installing.

Who agents name
category leader53%
runner-up18%
your tool7%
Stage 2 · Installed

Named isn't installed.

Agents name plenty of tools, but only install one. The gap never shows up in your analytics: the developer just ships whatever the agent picked.

Named → installed
namedinstalledleader 53%41%your tool 7%3%
Stage 3 · Compiled

A broken build sends the agent to someone else.

An agent installs you, then the build breaks on code written from outdated docs, and it quietly retries with a competitor. We run every build and record the result.

Successful builds · 39 of 50
Stage 4 · Kept

One “modernize this” prompt can delete you from a codebase.

Refactors are routine work for agents. Retention measures whether your integration survives one, or gets silently swapped for a rival or a self-hosted stack.

Kept after a refactor pass
kept94%
swapped out6%
Improve

We don't just measure. We move the number.

Every fix targets a failure we measured. Then we re-run the exact benchmark and put the before and after on the record.

01
Diagnose

Find where you lose agents

We read the failing runs and pinpoint where agents skip you, break the build, or drop you in a refactor.

02
Fix

Write what agents need

Corrected docs, examples, and SDK guidance, targeted at exactly the failures we measured.

03
Re-test

Watch it move

We re-run the exact benchmark, same prompts, same models, and put the before and after on the record.

Measure → Improveillustrative product example, synthetic figures
Build success84%
+27 pt after the fix
Retention98%
+4 pt after the fix
MeasuredAfter the fix

Agents are already picking winners. Find out if you're one.

Your engagement results stay private. Public research follows one published policy for everyone.