2026 Report · Irina Nazarova · Evil Martians · SF Ruby
CEO at Evil Martians
Co-founder at AnyCable
Org
of SF Ruby Community
Rebulk · YC S25
Rebulk
Rebulk
Cora
Cora
Cora
Same five founders, one more question: what would make Rails better? Their wishlist. ↓
It's landing: Falcon · fiber-aware Active Record · Action Cable re-architected
A fix? Ship the docs as agent skills.
Hiring or looking? sfruby.com/jobs
A fix? Inertia brings React's libraries to Rails — inertia-rails.dev
The front end was the one all five named. It can be live and interactive — Marco Roth is rebuilding it.
Let's jump ahead — an experiment: Rails in your browser, on PGlite. →
PGlite · ElectricSQL · ruby.wasm · palkan/wasmify-rails · on years of work from Vladimir Dementyev & Evil Martians
AI agents are handed real coding tasks, never told the language. The tasks are his, not ours. For the product builds: how many picked Ruby?
0 of 210 product builds. Not one.
GET /widgets-orm/42 · req/s, median of
5 Fly performance-16x hosts
(16 vCPU · Postgres 17 loopback ·
200 conc · oha) · 2026-06-10.
github.com/irinanazarova/small-runtime-bench
Cost for Claude Code to build the same project in 13 languages · Ruby is cheapest. Benchmark by Yusuke Endoh (mame) · dev.to
Ruby is fast and token-efficient.
Why aren't
models picking it?
Crawlable 85/92 · Sitemaps 49/92
llms.txt 43/92 · .md routes 21/92 · content negotiation 13/92
Of 12 common product builds, 0 have a solid, numbers-backed case for Rails
shared gem/agent convention
none yet · Agent Skills
fragmented
A start:
github.com/Bakaface/rails-hyperdrive
— an MCP server + skills for any Rails app
Ruby in Multi-SWE-bench absent · idiomatic-Rails dataset none yet
I re-run the benchmark every year. 0 of 210 is the number to beat.
It's up to us to make people and models know.