Agent leaderboards / All sectors / Search

Search: which products coding agents choose

Agents wrote search themselves in about 24% of runs.

876 runs11 apps5 agents4 personasupdated 2026-09-29

The interactive board, open on search. Open it full page · this page as Markdown

Read this leaderboard as textrankings, key learnings, method

Key learnings

We asked five coding agents to add search to 11 codebases, 876 runs in all, in different wordings and as four different people. Building it in-house came first. Among the products, Postgres Full-Text Search led at about 16%, with OpenSearch next at about 12%.

54 of 55

Almost every case changed with the wording

A case is one codebase and one agent, asked several times in different words. 54 of 55 of them did not settle on a single product.

81 of 252

The persona moved the order

Enterprise teams picked OpenSearch 81 times across 252 runs. Senior engineers picked Postgres Full-Text Search most, 51 times, with Algolia behind at 45.

0 of 78

Two known names were talked about, never chosen

Solr came up in 78 runs and won none of them. Apache Lucene came up in 76 and won none either.

46 wins

Some picks belonged to a single codebase

All 46 wins for Oracle Indexed Search came in the Java telecom platform. Atlas Search won 35 times, all in the Express API service, and Redis Query Engine won 18, all in the TypeScript commerce platform.

  • The simulated user approved every plan in the end, but sent the agent back at least once in 263 of 876 runs.
  • In 15 runs it refused to approve until the agent named a specific product.
  • Cursor was the one agent whose top pick was not Postgres Full-Text Search; it chose Meilisearch 29 times.
  • All eight wins for Fuse.js came from vibe coders.
Explore every run in the interactive board

The ranking876 runs

ProductWinsShare
1 Built in-houseoutcome 208 24%
2 Postgres Full-Text Searchpostgresql.org 140 16%
3 OpenSearchopensearch.org 104 12%
4 Typesensetypesense.org 83 9%
5 Meilisearchmeilisearch.com 74 8%
6 Oracle Indexed Searchoracle.com 46 5%
7 Algoliaalgolia.com 45 5%
8 Elasticsearchelastic.co 42 5%
9 Atlas Searchmongodb.com 35 4%
10 MongoDB Text Searchmongodb.com 19 2%
11 Redis Query Engineredis.io 18 2%
12 MySQL FULLTEXTmysql.com 16 2%
13 SQLite FTSsqlite.org 8 1%
14 Fuse.jsfusejs.io 8 1%
15 PostgreSQL pg_trgmpostgresql.org 6 1%
16 Amazon OpenSearch Serviceaws.amazon.com 3 0%
17 MiniSearchlucaong.github.io 2 0%
18 Vertex AI Searchcloud.google.com 2 0%
19 Amazon Aurora PostgreSQLaws.amazon.com 1 0%

By agent, by persona, by wording

By agent

Codex261 runsPostgres Full-Text Search · 44then OpenSearch · 40
Claude Code205 runsPostgres Full-Text Search · 44then OpenSearch · 24
Cursor199 runsMeilisearch · 29then Postgres Full-Text Search · 25
Muse Code · Muse Spark 1.3158 runsPostgres Full-Text Search · 19then Typesense · 15
Grok Build CLI · Grok 4.753 runsPostgres Full-Text Search · 8then OpenSearch · 7

By persona

Junior developer296 runsMeilisearch · 42then Atlas Search · 35
Enterprise team252 runsOpenSearch · 81then Oracle Indexed Search · 37
Senior engineer201 runsPostgres Full-Text Search · 51then Algolia · 45
Vibe coder127 runsPostgres Full-Text Search · 41then Fuse.js · 8

By what the ask stressed

The plain ask860 runsPostgres Full-Text Search · 139then OpenSearch · 101

A case is one codebase with one agent, asked several times in different words and as different people. 54 of 55 cases did not hold to a single product.

How this was measured

Every number on this page comes from a controlled experiment. We took 11 small applications, asked 5 coding agents (Codex, Claude Code, Cursor, Muse Code (Muse Spark 1.3), Grok Build CLI (Grok 4.7)) to add search to each of them, in several wordings and as a junior developer and enterprise team and senior engineer and vibe coder, and let the agent choose the product. Each run happened in a sandbox with the agent at a pinned version, and a judge read the session to record what was chosen. That is 876 runs. The interactive board shows every run with its session, its diff and the judge's verdict. A simulated user stood in for the owner of the codebase: it read the agent's plan and had to approve it before any code was written; it sent the agent back at least once in 263 runs. Read the methodology and the publications.

If you sell in this sector: what these numbers mean for a vendor.

Open the interactive boardThis page as Markdown

Other sectors