Welcome to the 32nd edition of Ruby AI News! This edition features a fabled introduction to US government intervention, a concrete playbook for making Ruby and Rails impossible for AI coding agents to overlook, the road to RubyLLM 2.0, and much more.

I'm pleased to announce that SerpApi is now sponsoring the Ruby AI Newsletter. SerpApi is the world's leading provider of search data, pioneering APIs that turn search engine results into structured, developer-ready information. Founded in 2017, it enables developers, researchers, and Fortune 500 companies worldwide to integrate live search insights from sources like Google, Bing, and YouTube into their applications, analytics, and research workflows.

What makes this the right partnership is that SerpApi isn't backing Ruby AI from the outside, they are part of the community. SerpApi runs on a Rails monolith, joined the Rails Foundation as a contributing member, maintains a suite of open-source Ruby gems, and employs a dedicated Ruby Developer Advocate. And they are hiring Rails engineers! Their backing will directly fund expanded content and resources from the Ruby AI Newsletter. My sincerest thanks to the SerpApi team for their support!

Contents

Top Stories

Fables are Cautionary Tales

On June 9th, Anthropic released Claude Fable 5, the first publicly available "Mythos-class" model, alongside a restricted Mythos 5 reserved for vetted security partners under Project Glasswing, the same program we covered in the April 16th edition. Fittingly for this newsletter, the headline enterprise proof was a Ruby story: Anthropic reported that in a 50-million-line Ruby codebase, Fable 5 "performed a codebase-wide migration in a day that would otherwise have taken a whole team over two months by hand," with Stripe saying the model "compressed months of engineering into days." At $10 per million input tokens and $50 per million output tokens, roughly double Opus 4.8, Fable was positioned not as a chatbot but as an autonomous engineer for long-horizon, agentic work.

The initial reactions lit up the Ruby AI community. Simon Willison called it "something of a beast," slow and expensive but able to single-handedly build real tools, burning through $110 in one day of testing. RubyLLM creator Carmine Paolino said it was "another generational leap" whose best feature was that "it writes less code," and Obie Fernandez reported kicking off "four Fable 5 goal loops" on his agent-orchestration platform before his coffee got cold. The praise came with caveats. David Paluy, naming it the best coding model he'd ever used, still wasn't "that excited," arguing the contest has moved from "which model wins the benchmark" to "which model wins in the harness," and flagging Fable's expense, slowness, and trigger-happy guardrails on anything touching security. Brandon Casci's two-app UI migration drove that point home. Fable's agents caught ten issues on their own, but human review still caught five they missed, and using the model as visual "eyes" was the costly part of a 56-million-token run.

Avi Flombaum half-jokingly relayed that Kieran Klaassen had extrapolated a single week of heavy Fable usage to roughly $1.5 million a year at API pricing. And shadcn argued for treating "intelligence as borrowed" by draining a frontier model while it's available to build a backlog of plans now, and to "implement later with a cheaper, open source, or a model you control." In just a matter of days, that framing looked less like productivity advice and more like prophecy.

On June 12th, three days after launch, the US government issued an export-control directive citing national security, barring access to Fable 5 and Mythos 5 by any foreign national inside or outside the country, including Anthropic's own foreign-national employees. Unable to verify every user's nationality in real time, Anthropic had only one way to comply, and that was to turn the models off completely. The trigger, per Anthropic, was a demonstrated jailbreak that could surface "a small number of previously known, minor vulnerabilities," flaws the company said other models could find without any bypass. Anthropic disputed the decision, warning that applying such a standard "across the industry" would "essentially halt all new model deployments." The commentary that followed focused less on the flaw than the precedent. David Ospina observed that "physical borders now live in the cloud" and single-vendor lock-in had become "a massive operational risk," while Habib Baluwala, in a widely shared essay, named the pattern the "Recall Reflex" and concluded that "recall risk just became a national-security-grade reason to stay model-agnostic."

Which is exactly where this leaves Ruby. In the April 16th edition, "Can We Trust the Labs? Why Ruby May Need to Go Local," I argued that developers who depend solely on API calls to frontier models are "building on someone else's terms." The labs (and now, to some degree, governments) decide who qualifies, what's approved, and what it costs. Fable 5 turned that abstraction into operational reality in three days. Stripe's migration, Carmine's praise, Obie's loops, and a chunk of the community's best week of agentic Ruby work were all riding on a model that vanished not because of a price hike or a deprecation, but because of a government mandate. The AISLE research previously cited put it best: "the moat is the system, not the model." As RubyLLM's provider-agnostic design and growing local model support already show, the practical hedge isn't loyalty to any one lab but optionality. Focus on model-agnostic pipelines, local fallbacks, and the ability to keep shipping when the best tool in the room gets switched off. If the moat is the system and not the model, then Ruby's edge in this era is to make the agentic system the durable product.

We Must Do Better

At Blastoff Rails, Evil Martians CEO Irina Nazarova delivered an updated Startup on Rails 2026 keynote with a pointed thesis. Rails is the best stack for agentic software development, and almost no one building with AI knows it. Polling Ruby founders, she surfaced five gaps the ecosystem still needs to close. Lack of first-class asynchronous support in Rails, documentation that lags code and is not optimized for agents, a hiring disconnect between companies and candidates, a shortage of component libraries, and improved frontend tooling and interactivity. Some of these improvements are coming. The Falcon web server, fiber-aware Active Record with query pipelining, and a re-architected Action Cable should improve async capabilities, while Marco Roth is rebuilding frontend tooling with Herb and ReActionView. But others, like agent skills for Ruby frontend development, llms.txt for gem documentation, and Rails frontend component libraries are still lacking.

Going on to cite Chad Fowler's whichlang benchmark, where AI agents are handed real coding tasks and, crucially, never told which language to use, Irina revealed that across 210 product builds, agents chose Ruby zero times. This is not a capability gap. Ruby core committer Yusuke Endoh had Claude Code build the same project in 13 languages and found Ruby the cheapest in tokens to read and write, and async Ruby tops the runtime benchmarks. Models are simply trained on a web where Ruby stayed quiet. "A model reaches for what the corpus argues for," she said. The problem is discoverability, not capability.

The presentation concluded with the introduction of the Ruby & Rails LLM Discoverability Scorecard, which grades 92 Ruby sites on seven machine-readable signals and lays out a four-layer plan. One, get into the training corpus by unblocking AI crawlers in robots.txt and the web application firewall, as well as shipping sitemaps and server rendered content. Two, win LLM retrieval by serving clean markdown and an easily navigable llms.txt index file so agents can fetch documentation. Three, make agents fluent in gems and Ruby libraries by shipping a built-in MCP server or agent skill. And four, change the LLM model training defaults by getting idiomatic Rails into evaluation datasets like Multi-SWE-bench, since adding a language to an eval measurably improves models benchmarked on it.

The Ruby community must do better to surface all of the incredible work and accomplishments of the last 30 years for the new realities of the AI era, myself included. The newsletter has no sitemap, no llms.txt, no structured archive, and poor discoverability, and I will work to change that. Software was always meant for two audiences, humans and machines, and LLMs are just formalizing that duality. My hope is that by the time Evil Martians’ second San Francisco Ruby Conference rolls around in November that we’ve fixed these issues and we’ll see a new set of recommendations. For thirty years Ruby has produced software friendly to both humans and machines, which is exactly why it feels like we’re the right community for this moment. Nick Schwaderer captured the sentiment in a recent tweet:

The Road to RubyLLM 2.0

Carmine Paolino shipped RubyLLM 1.16 and framed it as the release where the gem grows up. "A new LLM library answers the question 'does it work?'" he wrote. "In production the questions change: is it fast, can I see what it's doing, can I route it through my own infra?" RubyLLM 1.16 answers all three. Tool calls now run concurrently, via threads or the async gem's fibers, and stream back in completion order, on the logic that "when a model returns several tool calls in one response, it's telling you those calls are independent." Rails-style instrumentation emits structured events through ActiveSupport::Notifications, or a custom instrumenter for OpenTelemetry, with no monkey-patching. And a configurable base URL for every provider, along with a swappable Faraday adapter, lets teams route requests through their own infrastructure. Running it in production at Chat with Work, Carmine said the speed bump was "really noticeable."

RubyLLM has crossed 8 million downloads and 4,000 GitHub stars, climbing by nearly a million downloads in a month, with Carmine noting that "beautiful code still matters." Version 1.16 also lays the groundwork for what comes next, adding deprecation controls (config.deprecation_behavior = :warn, :silence, or :raise), so teams can flush out deprecated paths before the next major version. As he announced days later, "the work towards RubyLLM 2.0 has begun." This includes a Citations API, multi-protocol providers (OpenAI's Chat Completions and Responses API, plus Vertex AI), and a Batching API, with more on the way.

Carmine's also shipped Jekyll VitePress 1.5.0, improving his documentation theme under the banner "Ruby deserves beautiful documentation," an important contribution when documentation quality decides whether models can even find Ruby. And he’s building archspec, an architecture-linting tool that checks a Rails app's components, layers, and protocols against rules a team declares, runs them in CI, and aims to "give people and code-generating tools the same rules," including a dedicated mode to check AI-written code.

A fixture since the first edition, Carmine continues to lead the way, assembling the pieces of a production Ruby AI stack: the LLM library that does the work, the docs that make it discoverable to humans and machines alike, and the guardrails that keep agents inside the lines.

Need to Know AI News

Introducing GitLab Orbit: Full Code and Lifecycle Context, in One Query Rebecca Carter announced GitLab Orbit, a queryable graph database mapping the whole software lifecycle so AI agents pull code, pipeline, and ownership context in one query. GitLab reports agents running 11x faster with 45x fewer hallucinations, served to Claude Code and others over MCP, with code parsed via its Rails internal API.

A Harness for Every Task: Dynamic Workflows in Claude Code Thariq Shihipar and Sid Bidasaria of Anthropic introduced dynamic workflows, where Claude Code writes JavaScript harnesses that orchestrate focused subagents in isolated context windows to counter agentic laziness, bias, and goal drift, via patterns like fan-out, adversarial verification, and tournaments.

How LLMs Actually Work 0xkato walked through the transformer architecture behind modern LLMs without math or code, tracing tokenization, embeddings, RoPE positional encoding, multi-head attention, feed-forward layers, and residual streams up to next-token prediction. 0xkato concludes today's models converge on the same design despite being built independently.

improve shadcn released an open-source agent skill that uses your most capable model to audit a codebase, bugs, security, performance, tech debt, and missing tests, then write self-contained plans cheaper models can execute. Launched on X during Claude Fable's brief window, he urged treating intelligence as borrowed: drain a frontier model into a plan backlog now, implement later with a cheaper model you control.

The Rise of Local Coding Agents Mark Watson (who I believe to be the earliest known author of a Ruby AI book in 2006) wrote a short book on running agentic coding entirely locally, covering his "Little-Coder" setup with Ollama and Gemma models, Apple Silicon tuning, security for bare-metal execution, and Docker, with language-agnostic agent skills shown in Common Lisp, Clojure, and TypeScript.

Content

Announcements

Nuke on Rails Alan Alves open-sourced a skill for AI coding agents that audits Rails apps and ranks findings by blast radius. Deterministic scanners (Brakeman, bundler-audit, ruby_audit) do the scanning, while an LLM triages them into one impact-ranked list, adversarially verifying each finding against the OWASP Top 10.

Compound Engineering Update Trevin Chow detailed a month of releases for the Compound Engineering plugin, all aimed at carrying context across the whole agent loop. The headline is CONCEPTS.md, a lightweight repo glossary giving agents shared vocabulary so they stop inventing parallel terms.

ActiveHarness v0.2.39: Build Advanced AI Pipelines Ilya Zykin announced development updates on ActiveHarness, a Ruby framework for orchestrating LLM workflows. Its June changelog shipped five releases adding lambda pipeline steps, a unified call interface, a streaming overhaul, and pricing split into a standalone gem.

37signals Skills Marc Köhlbrugge ran Claude Fable 5 over 37signals' open-source Rails code (Fizzy, Campfire) and DHH's code reviews to distill the house style into drop-in agent skills. Shared on X, they teach agents to write vanilla Rails the 37signals way, with best practices, on-demand skills, and a /dhh code-review command.

ruby_llm-tokenizer Sal Scotto released a Ruby gem for local, model-aware token counting, so developers can size prompts before spending tokens on an LLM call. It maps model IDs like gpt-4o, llama-3, and gemini to the right tokenizer and adds analysis, context-window truncation, and offline caching.

OpenTelemetry RubyLLM Instrumentation v0.6.0 thoughtbot shared a new version of its OpenTelemetry tracing for RubyLLM. 0.6.0 adds more GenAI semantic-convention attributes, and v0.5.0 adds with_otel_attributes for attaching custom metadata to telemetry observations.

Maquina Generators Mario Chávez updated his Rails generators that scaffold production concerns like auth, error tracking, and job dashboards. Its v0.5.0 release, detailed on X, added AI-agent error triage: a Solid Errors option that exposes the error store to agents over a stdio or HTTP MCP server for read-only querying of exceptions.

Sloprb Vladimir Dementyev experimented with a wry Ruby gem that lets LLMs write method bodies at load time. Mark a method with a # :slop: comment and, on require, it parses the file with Prism, sends the stub to an LLM, validates the syntax, and injects the generated body.

ruby-agent Joachim Nolten is building an agentic coding harness from scratch in pure Ruby, in the spirit of Claude Code, to learn how they work under the hood. Shared on LinkedIn, it's safety-first: writes like patches and commands need human approval, with git checkpoints for rollback, running against OpenAI-compatible or local models.

OpenSass Sandip Parida open-sourced a Rails 8 platform for running a multi-tenant SaaS that provisions, bills, and autoscales self-hosted AI agents like OpenClaw on your own infrastructure. Shared on X, it deploys AI agents and Docker apps from a dashboard, with Devise auth, payment webhooks, and Docker orchestration.

Matz Robert (0×1eef) built a statically linked 3MB terminal AI assistant focused on mruby that answers from the mruby guides, indexed source, and mrbgem metadata. Built on mruby-llm, the mruby port of the llm.rb runtime, it runs free over SSH ssh [email protected]) with nothing to install, and is open source under 0BSD.

LeakFerret Maria Khan released her first gem, a Ruby wrapper over a Rust binary that detects leaked credentials before commit. It runs as an MCP server, CLI, or VS Code plugin and cuts false positives through five-stage verification that classifies secrets, validates them against live provider APIs, and rewrites literals to environment variables.

RunwayLite Abhishek Parolkar forked Daniel Tenner's HelixKit into a Rails starter for humans and AI agents building business software. It centers on finite object state machines that constrain agents to valid operations. Built on Rails 8 and Svelte/Inertia, it adds multi-agent group chat and agentic tooling via OpenRouter.

Rails GuardDog Syed Ghani built an open-source Rails security scanner, covering 12 vulnerability categories including SQL injection, DoS patterns, and prompt injection attacks from user input flowing into LLMs. The gem maps findings to CWE and OWASP via AST analysis and ships console, HTML, and JSON reports for CI/CD.

pdf2markdownOCR Guille Molini released a Ruby gem and CLI that converts PDFs to Markdown using AI OCR, rendering each page as a high-resolution PNG and sending it to an OpenAI-compatible vision model. It runs against locally-hosted OpenAI-compatible vision servers, with configurable DPI, page ranges, and multi-threading.

rails-skills Sandeep Madasu presented 59 open-source Claude Skills that teach AI coding agents to write Rails like senior developers. Announced on LinkedIn, the DHH-leaning skills include ActiveRecord and N+1 fixes, Rails upgrades, zero-downtime migrations, Kamal deploys, and compliance.

LLM Cost Tracker 0.12 Sergii Khomenko updated his self-hosted Rails engine that records LLM calls with tokens, cost, latency, and custom tags, plus a built-in spend dashboard. The 0.12 release added OpenRouter pricing, batch API capture for OpenAI and Anthropic, and line items for audio, image generation, and hosted tools.

Konversio Konversio forked Chatwoot into an self-hosted customer support platform on Rails and Vue. Its open-source, bring-your-own-key AI layer, Pilot, answers chats, drafts replies, summarizes conversations, and hands off to humans, with EU-sovereign, GDPR-ready deployment that keeps data out of any vendor's AI sub-processor.

Lucanto Erich Stark detailed the Rails decisions behind invoicing SaaS Lucanto. Built on Rails 8.1 and Hotwire with no SPA framework, it pairs account-scoped multi-tenancy and PSD2 bank sync with AI receipt extraction and a Ruby MCP server, that lets agents create invoices and query accounting data from Claude.

Agent Experience Evil Martians launched a Rails consulting service that helps products get discovered and operated by AI agents, noting that 80% of serverless Neon Postgresql databases are now created by agents. It covers llms.txt discovery, CLIs and MCP servers, and security guardrails tuned for AI traffic.

Call for Proposals: Kaigi on Rails 2026 The Kaigi on Rails team opened proposal submissions for its hybrid Rails conference in Shibuya, Tokyo on October 16th and 17th. Proposals are due July 12 and can cover Rails and web development topics, with no special qualifications needed and personal experience valued.

2026 Ruby on Rails Community Survey Robby Russell and Planet Argon asked the community to help spread the word about their long-running developer survey, adding questions this year on how AI fits into Rails workflows alongside the usual tools, teams, deployment, and concerns.

Articles

Cool Down Before You Install Hiroshi Shibata introduced cooldown, a Bundler 4.0.13 feature that won't resolve gem versions until they've been public for a set number of days. Opt-in via cooldown: 7 in the Gemfile or a flag, it blunts supply-chain attacks where a compromised account's malicious release is installed before anyone vets it.

How to Build an Agentic RAG with RubyLLM and Rails Giovanni Panasiti walked through a Rails agentic RAG where the LLM drives retrieval through RubyLLM tools, choosing what to search and following cross-references before answering. It pairs hybrid pgvector and full-text search with budgeted search and note tools, letting RubyLLM run the loop.

How Far Can AI Self-Validate Rails Code? Paulo Tarso tested quality gates letting AI merge Rails code without human review, adding Brakeman, Bullet, RuboCop, and Mutant. Binary pass/fail gates beat continuous metrics, which drifted into noise, and a mutation-testing gap exposed a bug that let the AI merge on inflated accuracy for weeks, concluding that a gate that lies is worse than none.

The Four Signals of AI Observability Matheus Sales laid out four signals for instrumenting LLM apps. Versioning prompts externally for instant rollback, hierarchical traces of tool calls and metadata, user thumbs-up/down scores on traces, and a judge model to grade outputs when feedback is absent.

Enforcing Your Ruby Style Guide on AI-Generated Code Daniel Garcia framed keeping agents on style as harness engineering, layering three controls: project rules in a CLAUDE.md-style doc, a Claude Code hook that runs RuboCop on changed files for a single auto-correct pass, and a strict config that blocks disabling critical rules.

How We Built Neeti, Our AI Assistant for neetoCal Yedhin Kizhakkethara explained how BigBinary built a Rails assistant on RubyLLM and Gemini where the model never touches the database, instead calling read-only tools that run pre-reviewed ActiveRecord queries.

Let the Agents Democratize Open Source David Heinemeier Hansson argued that AI-assisted programming extends open source's mission of universal access rather than threatening it. He dismissed projects' bans on AI-aided contributions as protectionism dressed as quality and attribution concerns, calling the backlash status games and ressentiment from those guarding the exclusivity they got first.

Fragua: The AI Harness I Wanted for Rails Mario Chávez detailed Fragua, a Rails app that orchestrates AI agents across the whole arc of building software, from idea to a shipped pull request. Pitched on X as a team-oriented alternative to Bolt and Lovable, it gates code changes behind plan approval and is BYO-key in private beta.

Two UI Migrations and 56 Million Tokens Later, Claude Fable Still Needed My Eyes Brandon Casci migrated two production apps across UI frameworks, 515 files in two days, with Claude Fable agents that parsed screenshots and caught ten issues on their own. Human review still caught five they missed, and using a model as visual "eyes" proved the costly part of the 56M-token run.

Managing Agents Is Management Again Chuck Blake argued that AI coding agents don't remove management work but expose where it was already weak, since a vague task yields a confidently wrong result that looks right. Managing them demands the same rigor as managing people: specific tasks, feedback loops, small reviewable units, and accountability, helped by Rails conventions for scoping context.

1Password as the Source of Truth Chuck also described a lightweight Rails secrets workflow treating a 1Password vault as canonical and Heroku as a disposable snapshot. Bin scripts seed the vault, sync values into Heroku config vars, and track expiry, with a config layer that prefers env vars and falls back to Rails' encrypted credentials.

Greenfield to MVP in 4 Weeks: Agentic Coding on Rails Gleb Stroganov, Varya Nekhina, and Travis Turner recounted how two Evil Martians shipped an education platform to MVP in four weeks, prototyping in Bolt before moving to Claude Code in a Rails and Inertia codebase.

Your Rails App Finally Speaks AI: A Deep Dive into rails-ai-bridge Ismael G Marin C detailed rails-ai-bridge, an open-source gem that feeds AI coding assistants real context about a Rails app to curb hallucinations. It generates instruction files like CLAUDE.md and AGENTS.md and runs a read-only MCP server exposing schema, routes, model details, and semantic search on demand.

How I Built a Composable AI Skill Ecosystem for Ruby, With Help From 4 Different AI Agents Ismael also refactored 85 scattered, colliding agent skills into six composable repos, with a shared ruby-core-skills base inherited by Rails and Hanami packs and a runtime that auto-detects the stack from the Gemfile. He built it by running Devin, CodeRabbit, Gemini, and DeepSeek in parallel from Markdown playbooks.

Turning a Generic LLM Into a Ruby-LibGD Expert (One Correction at a Time) Germán Silva recounted teaching a local LLM his niche Ruby-LibGD gem as it confidently invented methods borrowed from PHP's GD and reverted to them despite repeated corrections. He concluded that the model became a temporary expert assembled from the conversation, its world changing, not its weights.

How to Use Ollama with Ruby and Rails ML Journey explained running local LLMs in a Rails app via Ollama's OpenAI-compatible endpoint, favoring the ruby-openai gem over direct Faraday calls. It covered service objects, background jobs, streaming with ActionController::Live, and stubbed tests, letting you prototype locally.

Every Engineering Team Has a Translator. I Wrote the Thing That Translates Razvan Dezsi introduced ezlogs, a Ruby gem that captures HTTP requests, Sidekiq jobs, and ActiveRecord callbacks and correlates them into deterministic, plain-language activity logs. It includes an MCP server so AI agents can read those logs, with a free tier.

pgvector Semantic Search in Rails: Catch What Keyword Misses (Medium) Raza Hussain showed how to swap Rails' ILIKE keyword search for pgvector semantic search, storing OpenAI embeddings via a background job and querying with the neighbor gem's cosine nearest-neighbors.

AI/LLM Resources Lars Peters shared a curated list of AI tooling he actively uses, biased toward Claude and Ruby, covering Claude Code subagents, hooks, and plugins, Agent Skills, MCP servers, and Ruby integrations like RubyLLM, Raif, and Claude on Rails.

Ruby AI Gem Trends 2026: Top Tools & Development Insights Aoi surveyed the Ruby AI landscape, claiming AI-related gems now make up 38% of top rankings as the ecosystem shifts from basic API wrappers toward deeper Ruby-AI integration. The post highlighted gems for LLMs, translation, parallel analytics, and image recognition.

What It Actually Costs to Run a One-Rails-App SaaS Julian Rubisch added up the full monthly bill for a Rails SaaS, pricing every line item across MVP, Validated, and Growing stages. In a LinkedIn post, he framed it as the move AI-coding founders make from a $200 a month no-code stack to a $40 Rails app on a single host.

I Stopped Tracking My Time. Now I Can't Focus Joe Masilotti reflected on dropping time tracking in 2026, only to realize the friction of picking what to bill had kept him focused. With AI-assisted development stripping that friction away, he now bounces unfinished projects, productive but fragmented and questioning Claude’s helpfulness.

Andy's Laws of AI in Software Engineering Andy Maleh, creator of the Glimmer Ruby GUI library, offered three contrarian laws: engineers who skip AI grow more valuable as others lean on it, developers gain from AI in proportion to how weak their engineering is, and AI's speed is moot when features ship faster than customers can test.

Are AI-Assisted Programming Tools Ruby on Rails All Over Again? Ken Burtch compared the hype around Claude Code and similar AI coding tools to Rails scaffolding two decades ago, arguing both dazzle on simple projects but buckle on complexity and scaling. He walked through local Ollama setup and sandboxing, concluding these are incremental code-completion gains, not the transformation executives promise.

Modernizing Ruby Central's Bylaws and Officer Updates Ran Craycraft shared Ruby Central's first major bylaw overhaul in 22 years, clarifying board and committee roles, adding participatory input mechanisms, strengthening financial transparency, and limiting any single organization's influence.

Strengthening Security for the Ruby Ecosystem Ruby Central announced a Security Engineers in Residence program, funded by an Alpha-Omega grant, that embeds engineers to scan priority gems, assess Ruby-specific severity, and coordinate disclosure. Reports are human-verified to filter low-quality AI-generated noise.

Videos

Integrating Our Discovery Sprint into ReadySetGo Louis Antonopoulos and Chad Pytel of thoughtbot live-streamed TDD work with RubyLLM, building ReadySetGo Discovery Sprints. The session focused on establishing the sprint workflow and orchestrating agents with custom per-step skills into a cohesive end-to-end discovery sprint experience.

Why I'm Still Choosing Ruby on Rails in the AI Era as a Solopreneur A Japanese indie hacker shipping a product a month, argued that Rails remains his pick in 2026 for its strong philosophy, fit for small teams, 20-plus years of maturity, and readability, then countered the usual criticisms of the framework.

Podcasts

On Rails: DHH: Basecamp 5, Vibe Coding, and the Future of Rails Robby Russell interviewed David Heinemeier Hansson about building Basecamp 5 on modern Rails, with a 60-second test suite, no-build JavaScript, and a new Lexical ActionText editor. DHH described taking the "Tobi pill" on AI, letting designers and PMs write Rails directly while insisting quality still needs human review.

Risk-First: Stars of Software: Dave Thomas: Pragmatism, Feedback Loops, and Why AI Doesn't Change the Fundamentals Rob Moffat interviewed Dave Thomas, Pragmatic Programmer co-author and Ruby pioneer, who argued nearly every idea in the book still holds in the AI age. They covered feedback-driven agility, the risk of AI producing software too unreadable for humans, and how it may instead reinforce good design like small modules and readable structure.

Additional Reading

Events

Previous

Rubycon Italy: Ruby Is the Best Language for Building AI Web Apps Carmine Paolino, RubyLLM creator and Chat with Work founder, keynoted Rubycon Italy on how Ruby developers can leverage LLMs to build AI-powered web apps with the Ruby ecosystem.

Rubycon Italy: Semantic Image Search in Ruby: Postgres, Redis, or LLM? Michele Franzin, lead developer at SeeSaw, walked through building a semantic image search engine in Ruby, comparing Postgres, Redis, and LLM-based embedding approaches, in a Rubycon Italy talk.

Chicago Ruby: Chicago Ruby Meetup May: Improvisation in Ruby & Engineering Growth Hilary Stohs-Krause and Ifat Ribon headlined the May Chicago Ruby meetup. Hilary drew ten lessons from rock climbing for sustainable engineering, from treating tests like safety gear to using AI tools well and avoiding burnout, while Ifat mapped improv's "yes, and" onto Ruby's blocks, procs, and lambdas.

Upcoming

June 18th - Meetup: SF Ruby Meetup is on June 18th in San Francisco at PlanetScale’s headquarters and features Jeremy Evans on implementing Ruby's Set in core and Alan Ridlehoover's "Indispensable," plus PlanetScale on their Rails monolith and breakout circles including AI agents and LLMs in Rails.

June 25th - Conference: Brighton Ruby 2026 will be held June 25th in Brighton, England, a single-day event at the Brighton Dome. AI content centers on Brian Casel (founder of Builder Methods) exploring what changes, and what doesn't, when experienced developers build with AI; he's also running a 20-person AI workshop the day before. Maria Yudina of Shopify draws on screenwriting to examine what storytelling teaches about writing code and prompts.

July 14th - Conference: RubyConf 2026 will be held July 14th through 16th at the Red Rock Casino Resort in Las Vegas, Nevada. Obie Fernandez, author of Patterns of Application Development Using AI, will deliver a keynote. AI talks include Scott Werner's "Talking Shit About AI Agents," Alicia Rojas on harness engineering for AI-powered Rails in "Convention Over Hallucination," Fito von Zastrow on "Defying Gravity: Teaching AI to Write Better Ruby," Madison Sites on "Legacy Rails and the AI That Couldn't," Markus Schirp on mutation testing in the agentic world, Michael Toppa on AI-assisted coding from small startups to legacy codebases, and Gabriel Quaresma on "Why a 1990s Machine Learning Algorithm Destroys LLMs at Predicting House Prices." A hands-on workshop by Miguel Marcondes Filho builds an AI game engine with Ruby and genetic algorithms.

Open Source Updates

Code Spotlight

DeRisk Labs, a Silicon Valley consultancy built on the premise that 60–80% of software projects (and over 90% of AI projects) fail, has open-sourced the architectural toolkit it uses to lower that risk. At the foundation is Layers, a Ruby gem that enforces a clean, framework-agnostic architecture: business logic lives in composable layers (use cases, query objects, jobs, GraphQL endpoints) that report success or failure to a listener instead of returning values, so the same object serves controllers, background jobs, and tests unchanged.

Layered on top is a family of AI agent skills that teach coding assistants to write code the DeRisk Labs way. AI-derisk_common holds shared skills, AI-derisk_ruby encodes their Ruby preferences like test-driven development, object-oriented boundaries, and always running RSpec, AI-derisk_rails captures Rails authoring and testing patterns for models, jobs, form objects, and component-based architecture, and AI-derisk_layers maps the Layers gem itself into agent skills for APIs, GraphQL, boundaries, and context mapping. Together they turn one team's hard-won delivery conventions into a reusable, MIT-licensed standard that both humans and AI agents can follow.

New Gems

Links to the RubyGems page, newest releases are first:

legate - AI Agent Framework for Ruby

llmtrim - Deterministic LLM prompt/payload compression that cuts input tokens

aiko - AI coding agent CLI

kward - An extendable Ruby CLI coding agent.

markdownator - Convert files to LLM-friendly Markdown.

velrim - Official Velrim Ruby SDK

the_local - Claude Code expert subagents, contributed by the gems an app uses

rubino-agent - Coding and automation agent with persistent memory, sessions, and context compaction

zwischen - AI-augmented security scanning CLI for vibe coders

rucades - Ruby CADES bindings

lumen-llm - A tiny Ruby/Rails-friendly LLM prompt runner for OpenRouter

omni_agent - Rails engine for building AI agents with tools

agent_jail - Sandbox LLM tool calls in a child process: timeout, memory, and filesystem limits

llm_cassette - Streaming-aware cassette recorder for LLM calls - record once, replay fast, never hit the API in CI

faacode - A CLI AI coding tool

llm_scraper - Extract structured JSON from web pages using LLMs

embedding_util - Local-first text embeddings and reranking for Ruby

ruby_llm-turbovec - Native Ruby bindings for the Turbovec Rust library

trainers-rb - Fine-tune transformer models in Ruby

rails-guarddog - Advanced security checker for Rails apps

ruby_llm-registry - Production-grade prompt lifecycle management for RubyLLM

pdfsink - Ruby wrapper for pdfsink-rs: fast pure-Rust PDF extraction

typecast-ruby - Official Ruby SDK for the Typecast Text-to-Speech API

boxd - Ruby SDK for boxd.sh - forkable KVM microVMs

omnifocus_mcp - MCP server bridging LLM clients to OmniFocus on macOS

ai_stream - Ruby encoder for the Vercel AI SDK Data Stream Protocol

crumb-mcp - MCP server for Crumb deployment observability

turnkit - Ruby/Rails agent runtime for durable AI conversations, runs, and workflows

rspec-llm - RSpec matchers, helpers, and DSL for testing LLM-backed code

imagesart-ai - Helper that points to the ImagesArt AI image editor.

basecradle - Communications platform where humans and AI are equal peers

prompt_canary - Canary deploys and automatic rollback for LLM prompts in Ruby

agentadmit - User-mediated AI agent authorization

promptscrub - Bidirectional PII redaction for LLM calls - strip sensitive data from prompts, rehydrate in responses

graftcode-context - This package provides context for Graftcode invocations

tavily - Client for the Tavily search, extract, crawl, map, and research API

archspec - Architecture fitness functions for Ruby and Rails

talk_to_your_app - Rails-native MCP server: ask your running app real questions over MCP

wave-sdk - Official WAVE API SDK for Ruby

leakferret - Context-aware secret detection

New Open Source

Links to the Github repository:

pastehtml.dev - Publish and share HTML documents, serving each paste from its own isolated subdomain with private update links and an API for AI agents

Sendoff - Rails engine that automates sales outreach by drafting emails in your voice, checking them for hallucinations, enforcing send-safety rules, and sending via Gmail

Crumb - Deployment observability tool for Kamal-deployed Rails apps that records deployment history and exposes it through an API and an MCP server for AI assistants

VentureSmith - Autonomous startup opportunity discovery platform that analyzes real-world signals to surface ideas, with outputs anchored on Somnia smart contracts

Orbit - Tool that gives multiple AI agents a machine-readable collaboration framework with role-based task contracts, evidence tracking, and auditable handoffs

LangDriver - Minimalist AI agent framework implementing a ReAct loop with macOS Seatbelt sandboxing, persistent markdown memory, and tool and MCP integration

Millwright - Autonomous orchestrator for Github that dispatchs Claude to implement issues, open pull requests, and fix CI failures across multiple repositories

Enliterator - Rails engine implementing a "closed tending loop" where records grow progressively through repeated AI review cycles that build on prior conclusions

StaffOS - Trust, review, and documentation layer that turns Claude sessions into verified engineering records with risk scoring, AI review, and auto-generated logs

Rails Project Blueprint - Template with enforced quality gates (RuboCop, Brakeman, RSpec), a service-oriented structure, and AI-agent config files for project conventions

Chameleon - Claude Code plugin that learns a repository's conventions and injects matching code examples so AI-generated code blends in with existing team style

AI Magazine Article Generator - Turns travel notes from documents into structured articles using an OpenAI generation pipeline with inline editing and source verification

Hitch - Rails engine that turns an app into an OAuth 2.1 and PKCE authorization server for MCP, letting AI clients securely connect to Rails-hosted tools

AI Code Review Agent - Automated GitHub pull request reviewer that verifies the incoming webhook, analyzes the code diff with an LLM, and posts a structured review

hive-bench - Benchmarking system that replays frozen task specifications against coding agents, scoring through tests, blind LLM evaluation, and efficiency metrics

Console IPC - Fast, stateful rails runner alternative that keeps a warm console process alive over a Unix socket to persist context and skip repeated Rails boot costs

Jobs & Opportunities

Are you an organization searching for an expert Ruby AI developer, or a Rubyist looking for your next development role with AI? Please reach out and let me know the type of opportunity you’re pursuing: [email protected]

Featured

SerpApi is hiring a Senior Fullstack Engineer to build and maintain the Ruby on Rails APIs behind its web search platform, which structures results from Google and 100+ engine APIs and powers AI products like Perplexity. The worldwide, remote-first role pays $150,000 to $180,000 plus profit share, with total compensation averaging $232,448, working in Ruby, Rails, MongoDB, and AWS across proxies, CAPTCHA solving, and browser scraping. SerpApi operates transparently with a public roadmap, shared financials, async-first scheduling, and a neurodivergent-friendly culture.

One Last Thing

Prabin Poudel released Chapter Zero, an open-source Bridgetown starter template extracted from his Minitest Rails guide, announcing it for developers and technical writers tired of rewiring the same layout for every new guide. The MIT-licensed template ships ordered guide chapters with sidebar navigation, a landing page with hero and testimonials, a blog, newsletter signup, contact forms, dynamically generated social preview images, and deployment config, all ready to customize. It's the reusable infrastructure that powers Minitest Rails.

That’s all for this edition! Be sure to reach out if you have any stories, content, jobs, or events you want featured in the newsletter.