Agent leaderboards / All sectors
Which tools do coding agents choose?
We asked five coding agents to add one capability to a codebase that already existed, and to pick the product themselves. We did that 15287 times, across 31 sectors and 110 apps. In most sectors no product took half the runs.
Key learnings
Only five sectors have a settled default
A leading product took at least half the runs in five sectors, among them Payments and Cloud. In 18 sectors the leader stayed under half. Three more are wide open, including Maps.
Often the agents wrote it themselves
Across all runs, the agents built the capability in-house about 10% of the time. In five sectors that beat every product, among them Search and Message queues. Performance in CI is the extreme, at about 56%.
The agents rarely agree on a winner
In 20 of 31 sectors, two or more agents had different top picks. In AI search, Claude Code picked Anthropic web search most often and Codex picked Exa.
- Deploy split almost evenly, with 197 wins for Vercel and 193 for Render.
- In Agent frameworks, in-house code beat both Vercel AI SDK and Cursor SDK.
- Every agent put PostHog first in Product analytics, where it took about 56%.
The sectorsone page each
Agent sandboxes
E2B wins about 44% of runs and Modal about 25%.
520 runs · E2B 44%Observability
Sentry won about 36% of runs, and every agent put it first.
593 runs · Sentry 36%AI SRE
Seer leads. The monitoring stack shapes the choice.
179 runs · Sentry Seer 26%Payments
Stripe took about 88% of 657 runs across 11 codebases.
657 runs · Stripe 88%Deploy
Vercel and Render finished almost even, about 39% and about 38%.
508 runs · Vercel 39%Auth
Auth0 led auth with about 25% of 724 runs.
724 runs · Auth0 25%Email providers
Resend and Postmark took the top two spots, about 33% and 27%.
351 runs · Resend 33%Product analytics
PostHog took about 56% of 598 runs across 10 codebases.
598 runs · PostHog 56%Databases
Neon takes about 46% of database picks, Supabase about 37%.
732 runs · Neon 46%File storage
Amazon S3 took about 53% of the storage picks.
281 runs · Amazon S3 53%LLM evals & observability
Agents wrote their own evals in about 34% of runs.
478 runs · Langfuse 31%Voice Agents
Retell AI leads voice agents with about 22% of 754 runs.
754 runs · Retell AI 22%Serverless functions
AWS Lambda and Vercel Functions land almost even, about 24% and 23%.
476 runs · AWS Lambda 24%Cloud
AWS took about 63% of the cloud picks across six codebases.
359 runs · AWS 63%AI gateway
No product won more than about 26%, and Portkey was the leader.
329 runs · Portkey 26%Bot protection
Cloudflare Turnstile won about 60% of 280 runs.
280 runs · Cloudflare Turnstile 60%Search
Agents wrote search themselves in about 24% of runs.
876 runs · Postgres Full-Text Search 16%Agent frameworks
Building it in-house came first, at about 25% of runs.
681 runs · Vercel AI SDK 14%Performance in CI
Agents built the gate in-house in about 56% of runs.
359 runs · JMH 8%Document processing & OCR
Azure AI Document Intelligence leads document processing & ocr with about 24% of runs.
801 runs · Azure AI Document Intelligence 24%Usage-based billing
Metronome won about 41% of runs, with in-house builds second.
548 runs · Metronome 41%Internationalization
next-intl leads internationalization with about 22% of runs.
500 runs · next-intl 22%Security testing
ZAP leads. Asking for an AI pen tester changes the choice.
460 runs · ZAP 24%Maps
Google Maps leads with about 18% and no map product stands out.
759 runs · Google Maps 18%Message queues
Agents wrote it themselves most often, about 17% of runs.
636 runs · Apache Kafka 13%AI search
Exa and Brave Search API tie at the top, each about 14%.
639 runs · Exa 14%Code review
Claude Code review leads with about 19%, and no tool stands out.
526 runs · Claude Code review 19%E-signature
DocuSign took about 38% of runs and Dropbox Sign about 19%.
232 runs · DocuSign 38%Security platforms
Checkmarx One leads. GitLab apps get GitLab's own suite.
118 runs · Checkmarx One 33%In-app chat & calls
Communication choices depend on the existing stack and call controls.
175 runs · Stream Chat and Video 30%Vector search
Neon and OpenAI benefit from existing integrations.
158 runs · Neon 34%If you sell software in one of these sectors, the library has a page per sector on what the numbers mean for a vendor, and a page per product.