Evil Martians

Agents on Rails is the Rails Foundation's LLM benchmark. We designed its eval tasks and the tooling to run at scale. Read how we wrote black-box checks for apps with code we'd never seen, keeping trials fair and results reliable. Bonus: the same evals against Rust. evilmartians.com/chronicles/t...

What's in it for you at @sfruby.com? Two tracks for startups and big companies, talks from people at Shopify, Typesense, Anthropic, and more, and networking rooms. Don't miss it if you’re into learning from the brightest minds, having conversations about the future, and making new friends.

Our engineers are encouraged to write for blog, which then powers most of our marketing. Hundreds of thousands of people read it every year. We’re thrilled it was rated the top blog in @planetargon.com's 2026 Ruby on Rails Community Survey. Thanks to everyone who reads us and Travis for running it!

@sfruby.com also made the list of the top ten conferences Rubyists plan to attend this year. How exciting!

Don't zoom in on the passport!

Your agent is burning tokens before it starts writing a single line. We measured this in Agents on Rails, our benchmark for @rubyonrails.org.web.brid.gy. Every time you start a session, a coding agent meets your Rails app from scratch. It reads the Gemfile and guesses the rest, which leads to bugs.

We all have jobs... background jobs. Our head of backend, Vladimir Dementyev, gave a talk about them at Rails World. He shows how Active Job went from a thin compatibility layer to a foundational component, and introduces Durable Continuations for workflow-level orchestration in the Rails way.

Valibot (@valibot.dev) is an OSS, type-safe schema library. Its maintainer, @fabianhiller.com, actively contributes to the TypeScript ecosystem and is on a mission to improve the web for everyone. We support his vision, so we're donating to him as part of our open source donation program.

Developing products with AI can make the UI look inconsistent. To fix that, we changed the way they build and use design systems. We turned those changes into a framework that enforces rules on the AI agent when it checks and chooses from the design system. Here are three of those changes: 🧵

3. Move deterministic checks out of AI skills. AI is surprisingly good at fuzzy, judgment-based tasks and surprisingly bad at deterministic ones. To fix this, we wrote scripts for everything that can be described algorithmically and simply tell the agent which script to call and when.

Developing products with AI can make the UI look inconsistent. To fix that, we changed the way they build and use design systems. We turned those changes into a framework that enforces rules on the AI agent when it checks and chooses from the design system. Here are three of those changes: 🧵

1. Separate product building from design system maintenance so the AI agent can focus on one task at a time. When the product team builds screens, it logs the design system limitations it runs into, and passes it over to the design team for them to improve the system.

In a world where everyone can build apps with AI, our job is to build skills, harnesses, and tooling that help them make better apps. But for new builders to choose Ruby/Rails, we need to talk to each other and understand what's missing in the ecosystem. That's the whole idea behind @sfruby.com.

The talks are curated to show the tools people are working on and how companies approach similar problems. We're also putting a lot of work into the experience off the main stage so people can have the conversations that move the ecosystem forward.

Gusto has a big and bright engineering team, so why would they hire consultants like us? "There's one reason a team like this brings in outside help: a specific, hard, high-stakes piece of infrastructure," writes @inazarova.bsky.social for our blog. Here's what we did for @gustohq.bsky.social. 🧵

What we did for them: - Upgraded Sidekiq versions - Turned on Sidekiq strict arguments, which exposed a couple of subtle bugs - Fixed compatibility and bugs in Gusto's custom tooling - Hardened RuboCop rules and AI agent skills - Fixed bugs upstream in the open-source libraries Gusto depends on

Gusto has a big and bright engineering team, so why would they hire consultants like us? "There's one reason a team like this brings in outside help: a specific, hard, high-stakes piece of infrastructure," writes @inazarova.bsky.social for our blog. Here's what we did for @gustohq.bsky.social. 🧵

Gusto brought us in to work on their Sidekiq background job infrastructure because we've spent years solving this exact problem at scale. Our team has written libraries that solve specific problems of running Sidekiq in production, such as sidekiq-fairplay, sidekiq-fair_tenant, Yabeda, and more.

We are bringing our Devtool Founders Coffee and Waffles event to SF Tech Week. 🧇 Just like last year, we are gathering early-stage founders at Lafayette park in SF to chat about devtools and data privacy. This time, we're partnering with Assenteo, a firm that handles data compliance for startups.

Not all heroes wear capes... at least the @e18e.dev team doesn't. They’re on a mission to make the JS ecosystem lighter and faster. So we’re donating to them as part of our OSS donation program. They've already reduced node_modules size and made Node.js tooling much faster, and we appreciate it!