Ruby vs everyone: the language that outlived its hype — and why I’d still start a product in it

Elena Vasquez

Elena Vasquez

September 18, 2026

Ruby vs everyone: the language that outlived its hype — and why I’d still start a product in it

I have been told Ruby is dead in at least four different decades of internet time. The first time I believed it was around 2016, when every new teammate wanted Node and every conference talk treated Rails as a punchline. I was writing Go services then. I was also quietly keeping a Rails app alive that made more money than the Go rewrite we never finished.

I would still start a certain kind of product in Ruby. Not a packet processor. Not a GPU trainer. A product: accounts, billing, admin, a pile of HTTP, a background job that must not double-charge Stripe. That is still most software that pays salaries. Here is why I would pick Ruby for it in 2026 — and the cases where I would not, even if I like the language.

What “vs everyone” actually means on a blank repo

The competitors I actually consider for a new product, not a tweet:

  • Rails vs Laravel vs Django. The honest peer group. Full-stack frameworks with batteries, an ORM, a story for jobs, and a culture of shipping a monolith first.
  • Rails vs Next.js plus something. The default I get offered by frontend-heavy teams. It is two runtimes, two package managers, and a BFF you will invent by week three.
  • Rails vs Go or Rust. I love both for CLIs and for services with a tight latency budget. I do not love them as the first place I build a form that needs validations, mailers, and an audit log.
  • Rails vs Node (Nest, Adonis, or a pile of Express). Same HTTP, worse defaults, and a culture that will talk you into microservices before you have a user.

Ruby the language is not the product. Rails is the product. Sidekiq is the product. Active Record, for all the N+1 jokes, is the product. Hotwire is optionally the product. When people say Ruby is dead they usually mean “I do not see it on Hacker News.” I mean “can I take a payment and send a receipt before lunch.”

Last year I started a small B2B billing add-on in Rails 7.1, then 7.2, then 8 on a Saturday because the upgrade was boring in the good way. Postgres, Solid Queue on one deploy, Sidekiq on another when I needed Redis anyway for caching. Devise for the first month, then I ripped it out for a thinner auth once I understood the session. Stripe via official gem plus a webhook controller I wrote myself because the abstractions always lie about idempotency. I billed a customer on day nine. I have not done that in a greenfield Go repo without lying about what “done” meant.

Developer at a cafe laptop working on a product dashboard

The language traits I still use on purpose

Ruby is slow if you write a tight loop like it is C. I do not. I write a loop that talks to Postgres and Stripe. The database wins that race. When I need a tight loop I have extracted it to Go or to a Rust crate and called it. That is a normal 2026 shape: a Rails app with a small foreign worker. It is not a betrayal of the language. It is knowing what the language is for.

What I still want from Ruby:

  • A readable domain layer. invoice.void!(by: current_user) is a sentence. I can sit with a finance person and walk the method. I cannot do that as easily with a 40-line Go function that returns (Invoice, error) six times.
  • Metaprogramming I mostly refuse, except where Rails already spent it. I do not write my own DSLs anymore. I use the ones that shipped: Active Record callbacks sparingly, Active Job, Active Model validations. The 2012-era “I invented a gem that rewrites your classes” culture is the part that deserved to die.
  • irb / pry as a product tool. I still debug production-shaped problems in a console against a staging database. The Java and Go versions of this exist. They are worse. A Rails console is why I can answer “what does this customer actually have” in two minutes.
  • Gems that are boring. pg, redis, sidekiq, pagy, view_component or plain partials, faraday or httparty when I must. The gem explosion of 2014 taught me to prefer stdlib and one well-kept library. That lesson applies everywhere. Ruby just learned it in public.

Sorbet and RBS exist. I have used Sorbet in an app that got big enough to hurt. I would add types at the boundaries — money, IDs, webhook payloads — before I typed every Active Record model. The language will not become Rust. That is fine. I did not pick it to become Rust.

Where Rails still beats the fashionable stacks

Convention is a feature when you are one or three people. I know where the mailer lives. I know where the job lives. I know that config/routes.rb is the map. A new hire who has done one Rails app can find the invoice path on day one. A new hire in a boutique Node repo has to ask which of the four “patterns” we meant this quarter.

Hotwire — Turbo and Stimulus — is the part I would actually bet on for a lot of UI. I do not need a Next.js app to update a table after a Sidekiq job finishes. I need a Turbo Stream and a small Stimulus controller. When I need a rich widget I drop in a Stimulus-wrapped Preact island or a React component via a single entry. I have done that. It is less glorious than a RSC architecture diagram. It is also less likely to become a second product.

Kamal (the successor to the “deploy from a laptop” dream that Mrsk started) is good enough that I have shipped to a Hetzner box without inventing Kubernetes. Docker, a load balancer, rolling restarts. For a product that is not a platform, that is the right amount of ops. I have also deployed Rails to Fly.io and to Render. I have deployed the same mental model to ECS when a customer required it. The framework does not care. The “Ruby cannot scale” slide always forgets Shopify, GitHub, and the boring company that just needed vertical scale and a read replica.

Jobs are first-class. Sidekiq is still the queue I trust. GoodJob and Solid Queue are good enough when I want fewer moving pieces. I have operated Celery and I have operated Bull and I have operated Temporal. Temporal wins when the workflow is the product. Sidekiq wins when the product is a web app that sometimes retries a webhook. Most products are the second thing.

When I would not start in Ruby

I would not start a latency-sensitive API edge in Ruby. If the pitch is “p99 under 20ms at the edge,” I want Go, Rust, or a managed edge runtime. I would not start a data plane, a game server, or a mobile client in Ruby. I would not start a team whose hiring market is 100% TypeScript and then surprise them with a Rails app unless I am ready to teach and to lose some candidates.

I would not start in Ruby if the UI is the product and the team is already excellent at React. A Figma-class canvas or a collaborative editor is not a Rails problem with a sprinkle of Stimulus. Use Rails as the API if you want the admin and the billing, and be honest that the editor is another app. I have seen the “it’s all Hotwire” dare fail. I have also seen the “it’s all Next” dare invent a worse admin. Split on purpose.

I would not use Active Record as a graph query engine. If the product is a social feed with twenty joins, I will write SQL, or I will use a tool that is for that. The N+1 problem is real. strict_loading, Bullet, and a habit of reading the log are the mitigation. Pretending the ORM is the villain while writing an ad-hoc query builder in Node is not a strategy. It is a change of scenery.

Server rack and laptop on a desk in a small office, warm evening light

The “Ruby is dead” cycle, named

Node was going to kill it because JavaScript is isomorphic. Then Elixir was going to kill it because Phoenix and the BEAM. Then Go was going to kill it because goroutines. Then Rust was going to kill it because the compiler. Then TypeScript everywhere was going to kill it because one language. Each of those languages is excellent at something. None of them made CRUD plus jobs plus email a solved problem better than Rails for a small team that wants to stay small.

What actually hurt Ruby was not a language. It was a hiring fashion and a conference fashion. Bootcamps moved to JS. Thoughtworks radar slides got mean. CIOs heard “legacy.” Meanwhile the apps kept taking payments. Shopify kept hiring. The Rails world got quieter and more adult: fewer magic gems, more explicit code, a framework that shipped Hotwire instead of pretending the future was a SPA in the asset pipeline.

YJIT and later JIT work made MRI less of a punchline. JRuby still exists for the shops that need it. I stay on MRI. I upgrade Ruby like I upgrade Postgres: on purpose, with a test suite, not as a personality.

How I would start tomorrow morning

If I were starting a product tomorrow that is a B2B web app with a dashboard, I would do this:

  1. Rails 8, Postgres, the default stack. Skip the kitchen sink. Add Sidekiq when I have a job that cannot be a perform_later in Solid Queue, or when I already need Redis.
  2. Hotwire for the first UI. A design system of plain partials. No Tailwind-religious war; I use it when I am the only frontend. I use a small CSS file when I am not.
  3. Stripe, webhooks, idempotency keys in a table. A Billing::ProcessWebhook that I can run from the console.
  4. A smoke test that signs up, starts a subscription in test mode, and sees the right entitlement. System tests are slow. One of them that hits money is cheaper than a rewrite.
  5. Kamal to a single VM until the VM is the problem. Observability: exception notices (I have used Honeybadger and Sentry), and Postgres query logs before I buy a second APM.

I would write the domain in boring Ruby objects when Active Record gets chatty — a Subscriptions::Start service, not a callback chain. I would keep callbacks for the truly incidental: touch a timestamp, enqueue a mailer. I would not build a “clean architecture” folder that maps every Rails concept into a Java package. I have seen that in Ruby. It is how you pay for two frameworks.

The comparison I give friends who are choosing

If you and your cofounder already live in TypeScript, start in TypeScript. The language war is not worth a six-month tax. Use a real framework — Adonis, or Laravel if you will cross to PHP, or Rails if one of you already has the scars. Do not start in a custom Express castle.

If you are choosing from a blank skill slate and the product is a business web app, I would still teach Rails first. You will learn HTTP, SQL, jobs, and mail. You will ship. You can learn Go when a service deserves Go. You can learn React when a screen deserves React. Starting in the language that is “more employable on Twitter” is how you build a toolchain instead of a product.

If you need a job at a FAANG-shaped company, learn the stack they list. I am not writing a career-advice piece dressed as a language piece. I am saying that for the act of starting a product — the act I have actually repeated — Ruby plus Rails is still the fastest path I know from empty directory to a customer who can log in and pay. The hype left. The loop did not.

I will keep writing Go for the sharp edges. I will keep TypeScript for the thick clients. I will still open a Rails app when I want the business to exist on Friday. That is not nostalgia. That is a preference I have re-earned every time a rewrite promised simplicity and delivered a second inventory of the same bugs.

More articles for you