Install share
The percentage of sessions in a category where a coding agent picked your product and wrote working integration code. The agent-era equivalent of market share.
Install share is the percentage of sessions, in a product category, where a coding agent picked your product and wrote working integration code.
It is the agent-era equivalent of market share, and it is deliberately strict about what counts.
What counts
Three conditions, all required.
The agent chose. Not listed your product as an option. Chose it, and moved on to implementing it. An agent that says "you could use Redis, Memcached or Valkey" has chosen nothing, and that session counts for nobody.
The code was written. A package in the manifest, an import, a call. If the session ended with a recommendation and no code, it is not an install.
The task was in your category. The denominator is sessions where an agent had to solve a problem your product solves, not all sessions.
Why it is the number that matters
Every other metric in this area is a proxy for it.
| Metric | What it is a proxy for |
|---|---|
| Brand mentions in AI answers | Whether a person hears about you |
| Citations | Whether a page of yours was read |
| Documentation traffic | Whether somebody looked |
| Install share | Whether you were bought |
For a developer tool, the last row is the only one that has already turned into revenue.
What a number means depends on the category
An install share is meaningless without the category's concentration.
| Category | Leader's share | What 10% means there |
|---|---|---|
| Payments | 88% | A strong challenger position |
| Databases | 64% | A real second tier |
| Authentication | 25% | Mid-pack |
| Agent frameworks | 14% | Close to parity with the leader |
Those figures come from 15,287 judged sessions across 31 categories.
Always split it
A single install share averages across markets that behave differently.
- By agent. Claude Code, Codex and Cursor did not all name the same category leader in 18 of the 29 categories all three ran.
- By kind of app. Requests to make an app's data persist went to one product in 280 of 359 sessions on apps with their own server, and to a different one in 111 of 142 on apps without one.
- By persona. The leader changed with the kind of person asking in 24 of 31 categories.
- With a rival installed. A written instruction naming a preferred product lost 2 times out of 3 once a competing library was in the repository.
A blended number can say you hold 22% of a category while hiding that you hold 40% with one agent and 5% with another. Those two situations need different work.
The denominator that people forget
Some sessions end with no product at all, because the agent wrote the implementation itself. That share, the in-house rate, ranged from 0% to 56% depending on the category in our data.
Those sessions belong in the denominator. If you exclude them, your install share looks better and you lose sight of your largest competitor.
Common questions
What is install share?
Install share is the percentage of sessions, in a product category, where a coding agent picked your product and wrote working integration code. It counts installs, not mentions or citations.
How is install share different from share of voice?
Share of voice counts how often a brand appears in generated answers a person reads. Install share counts how often an agent actually installed the product. One ends in a sentence, the other in a lock file entry.
What is a good install share?
It depends entirely on the category's concentration. In payments the leader holds 88%, so 10% is a strong challenger position. In agent frameworks the leader holds 14%, so 10% is close to parity.
How do you measure install share?
Run real coding agents on real tasks inside repositories that resemble your users' projects, with enough replicates to see past noise, and record what gets installed. Split the result by agent, language and persona.
Where this comes from
Armature ran 15,287 judged sessions with Claude Code, Codex, Cursor, Grok Build CLI and Muse Code inside 110 realistic codebases, and published every run. The numbers on this page come from that work.