标签：#rails

Scalable Guardrail Service ASP.NET Core Kubernetes: Architecture, Code, and Ops Quick Answer Scalable Guardrail Service ASP.NET Core Kubernetes: A dedicated ASP.NET Core guardrail microservice on Kubernetes validates LLM requests, enables instant policy updates via Redis, and scales with custom HPA for high‑throughput. Scalable Guardrail Service ASP.NET Core Kubernetes: Why a Dedicated Guardrail Microservice Matters When you expose an LLM‑powered API to the world, every request is a potential compliance risk. A single malformed prompt can surface PII, trigger a policy violation, or even cause a brand‑damaging output. In my experience, the first version of such a system is a set of ad‑hoc filters sprinkled across controllers. Under load, those filters become latency bottlenecks, policy updates race, and audit trails vanish. The root cause is a missing architectural layer that treats guardrails as a first‑class microservice that can scale horizontally, be updated live, and be observed independently. Guardrail Layer Requirements We need a guardrail layer that: Validates every request before it hits the LLM engine. Can be updated without redeploying the entire API surface. Provides per‑tenant isolation and versioning. Logs every decision for compliance and red‑team analysis. Runs at the same scale as the LLM inference service. When This Fails in Production Policy updates are applied via a shared ConfigMap and the pods do not reload, so new rules are never enforced. The guardrail service is single‑instance; a spike in requests triggers a queue that exceeds the LLM engine’s rate limit, causing a cascading failure. Audit logs are written to local disk; a pod crash loses events. Latency spikes because each request performs a synchronous Redis lookup for every policy. Common Mistakes Engineers Make Embedding guardrail logic inside the API controller rather than a dedicated middleware. Using in‑memory policy caches without a TTL, leading to stale rules. Ignoring the fact that

When I started studying APIs more seriously, I realized there was a problem with the way I was learning. I knew how to create a Rails API. I knew how to write: resources :products I knew what GET , POST , PATCH and DELETE were supposed to do. But I wasn't always able to explain why things worked the way they did. So I decided to go one step back and review the fundamentals: routing, HTTP, REST and how Rails puts all of these things together. This is what I learned. Rails Routing At its simplest, routing is the thing that connects a URL to some code in your application. In Rails, this happens in routes.rb . For example: get '/about' , to: 'pages#about' If someone requests: GET /about Rails knows that it should call: PagesController #about Pretty straightforward. But Rails gets much more interesting when we start using RESTful routes. resources does a lot of work Instead of manually defining every route for a resource: get '/products' , to: 'products#index' get '/products/:id' , to: 'products#show' post '/products' , to: 'products#create' patch '/products/:id' , to: 'products#update' delete '/products/:id' , to: 'products#destroy' Rails lets us write: resources :products And generates the conventional CRUD routes for us. HTTP Verb Action Purpose GET index List resources GET show Show one resource GET new Form for a new resource POST create Create a resource GET edit Form to edit a resource PATCH update Update a resource DELETE destroy Delete a resource This is one of the reasons Rails feels so productive. The framework isn't just giving us routing functionality. It is encouraging a convention. resource vs resources This one confused me for a while. resources represents a collection: resources :products There can be many products, so Rails generates an index route. resource represents a single resource: resource :profile There isn't an index because we're talking about one profile. It is a small difference, but it makes sense once you think about the resource you're mo

Fixing "g++ Not Found" When Debugging Rails in RubyMine on Fedora TL;DR: If clicking "Debug" in RubyMine on Fedora fails to install debase because of a missing g++ compiler, the standard @development-tools group might not be enough. Run sudo dnf install gcc-c++ make redhat-rpm-config to get the explicit C++ compiler and tools Ruby needs to build native extensions. I was recently working on a Rails app on my Fedora machine and wanted to step through some code. I fired up RubyMine, set my breakpoints, and clicked the "Debug" button, fully expecting everything to just work. Instead, RubyMine tried to automatically install the debase and ruby-debug-ide gems—which it needs under the hood to hook into the Ruby process—and threw a massive wall of error text at me. The core of the failure looked like this: Building native extensions. This could take a while ... ERROR: Error installing debase-3.0.17.gem: ERROR: Failed to build gem native extension. ... /path/to/extconf_common.rb:80:in 'Kernel#`' : No such file or directory - g++ ( Errno::ENOENT ) My system was essentially complaining that it couldn't find g++ , the C++ compiler. The Initial (Failed) Attempt My first thought was, "Oh, I must have forgotten to install the base build tools on this machine." Since I'm on Fedora, I reached for the standard DNF command to pull in the development group: sudo dnf install @development-tools (Note: If you're on older versions of Fedora, you might be used to dnf groupinstall "Development Tools" , but DNF5 uses the @ syntax or space-separated group install ). It downloaded and installed a bunch of packages. I felt confident, went back to RubyMine, clicked "Debug" again, and... got the exact same No such file or directory - g++ error. Why Didn't That Work? When you install Ruby gems that contain native C or C++ extensions (like debase ), Ruby doesn't just download a pre-built binary. It actually compiles the raw source code down to machine code directly on your machine so it runs as fast

When a build breaks, the bug fixes itself We stopped babysitting CI failures. Now a red build files its own bug — and an AI agent picks it up and ships the fix. PROBLEM — A failed build told no one Our CI would fail, and then… nothing would happen. The failure sat quietly in a build console that nobody keeps open. Eventually someone would notice a change hadn't gone out, go digging, and realize the build had been red for hours. And noticing was the easy part. Actually resolving it meant a whole code session: pull up the logs, find the failing step, reproduce it, and have an engineer sit down and personally shepherd the fix from broken to green. Every red build cost real human hours — plus the invisible tax of the delay before anyone even knew there was a problem. The true cost of a broken build was never the build. It was a person having to find it, understand it, and hand-fix it. SOLUTION — The failure files its own ticket — and an agent takes it from there Now nobody watches a console and nobody triages. The moment a build fails, it automatically files a bug in Shipeasy — our ops platform — as a real, prioritized ticket with the failing step, the branch, and a link to the logs already attached. From there it leaves human hands entirely. Shipeasy hands the bug to an AI agent, which investigates the failure, writes the patch, and opens a pull request against it. The loop that used to be "human notices → human reads logs → human fixes" is now "build fails → bug appears → agent fixes." The engineer's job shrank to reviewing a PR that already exists. DESIGN — How the whole thing hangs together The pipeline is deliberately boring — every hop is either something the cloud already does for free, or a service we already run: Cloud Build — build fails: a red deploy on main publishes automatically Pub/Sub topic → push subscription: filters to FAILURE · TIMEOUT · INTERNAL_ERROR HTTPS POST /webhooks/cloud_build Webhooks::CloudBuildController: verify token · decode · dedupe by

Hello! It has been a couple of months since I announced my latest project, Testing Rails from Scratch: A practical, (mostly) out-of-the-box approach to test-driven development in Ruby on Rails . The book is a thought experiment of sorts, as I revisit the default Rails testing stack after years of RSpec. I've been incrementally exploring the defaults and building tests from the foundational building block, on to more complex test cases and tools. This is the same approach I took to learn RSpec to begin with, and the same approach I took in the RSpec book on which Testing Rails from Scratch is based. It's been slower-going than I'd planned due to unexpected life circumstances this summer, so I wanted to share what is available now, and what's left. Quick plug for anyone still holding off: I'm running a work-in-progress special, $9 for lifetime updates to Testing Rails from Scratch until I'm finished. Purchasing work-in-progress books is a wonderful way to support independent authors, just saying! What's there now As I write this, the first five chapters of the new book are available for purchase on Leanpub. Chapters 1 and 3 are available as a free sample/preview download, no strings attached. An introduction to the problem being solved and my approach to learning and teaching Exploring the default Rails testing setup Testing Active Directory models Addressing DRY in tests, and when not to Using fixtures to create and manage test data Each chapter builds off the concepts and code from the previous one, and full source code for each chapter is provided. What's to come One interesting thing about my approach is I sometimes learn things as I'm writing and refining that affect future chapters. Case in point, I'd planned a single chapter on test data, decided to split it into two smaller, more digestible chapters. As of now, the remaining material will be broken up as: Managing test data with factories Testing units together with integrations Testing end-to-end with system

This week's Java roundup for August 3rd, 2026, features news highlighting: JEP 535, Shenandoah GC: Generational Mode by Default, targeted for JDK 28; point releases of A2A Java SDK, Apache Camel and Gradle; a maintenance release of GlassFish; the fifth milestone release of Groovy 8.0; and a follow-up of the JetBrains TeamCity CVE. By Michael Redlich

A multi-agent system without an orchestrator is just a collection of agents. Each one is capable, but none of them coordinated. They might all be excellent at their individual jobs - searching the web, writing code, calling APIs - but without something deciding what gets done, in what order, by whom, and what to do when a result comes back wrong, the system does not behave like a system. It behaves like a group project with no project manager. The orchestrator is the project manager. Its job is not to do the work. Its job is to make sure the work gets done - and that is a harder, more subtle problem than it sounds. What an orchestrator is responsible for An orchestrator does four things, and only these four things: 1. Decompose the goal. Turn a high-level objective into a concrete set of subtasks. This is a planning problem, not an execution problem. The orchestrator decides what needs to happen, not how to do it. 2. Route tasks to the right workers. Match each subtask to an agent capable of doing it. This requires knowing what tools and capabilities each worker has - not in detail, but well enough to delegate correctly. 3. Manage state across the workflow. As workers return results, the orchestrator decides what those results mean for the remaining plan. Sometimes a result changes the plan entirely. Sometimes it confirms the next step. The orchestrator holds the full picture. 4. Synthesise the final output. Worker outputs are partial. The orchestrator assembles them into a coherent response and decides when the goal has been met. Notice what is absent: the orchestrator does not call APIs, does not run code, does not search the web. It reasons about work and routes it. The moment an orchestrator starts executing, it loses the focus that makes it good at coordination. Building one from scratch Here is a minimal orchestrator in Python. It plans upfront, delegates to type workers, and synthesizes results: import json def orchestrator ( goal : str , workers : dict ) ->

Wide Events is a Rails gem that puts the production context a coding agent needs onto one OpenTelemetry root span per request or job. In one production search request, the root event showed 30.0 seconds total duration, 446 ms of Postgres time, and 29.4 seconds of outbound HTTP time. That was enough to focus the investigation on an external dependency. The trace then identified a POST that took 28.9 seconds. The trace contained 82 spans and 20,261 bytes of attribute JSON. The root event contained 40 attributes and 1,420 bytes. This is not a token benchmark, but it shows why the root event is a more compact starting point for an agent. Agents can read the code, but not the running system A coding agent starts with an unusual advantage: it can search every model, controller, job, migration, and test in a few seconds. It also starts with a serious blind spot. The repository cannot tell it: which account experienced the problem which build was running which feature-flag variant was active how many queries the request issued whether a semantic-search leg degraded how much an LLM call cost whether the same symptom appears in one tenant or every tenant Those answers often exist somewhere, but “somewhere” might mean a trace waterfall, application logs, a feature-flag service, product analytics, and a database console. Pulling all of that into a context window is expensive and usually requires several joins that were never designed in advance. A wide event changes the starting point. The app accumulates the context it learns while processing one unit of work, then attaches the completed flat map to the OpenTelemetry root span. The span is marked main=true , so every request or job can be queried as one row. request or job -> Rails and domain context accumulate -> child spans contribute dependency counts and timings -> one flat map is flushed onto the root span -> ClickHouse stores one queryable row The trace still exists. Wide Events gives it an application-shaped index. If y

I built DBT Companion, a free, mobile-friendly web app for exploring skills commonly taught in dialectical behavior therapy: https://dbt-companion.org The main product constraint was brevity without making the material vague. Each skill combines a short explanation with steps someone can follow. Users can also save favourites and use a browser-based diary card. A few implementation priorities were: making the core content usable on small screens keeping navigation predictable making privacy language visible rather than burying it keeping the stack simple with Rails, Hotwire, and Tailwind Diary cards are saved only in this browser's cookies. If cookies are cleared, saved diary cards will be lost. The app is educational, not a replacement for therapy, crisis care, or professional support. I'd welcome feedback on the accessibility, mobile interface, or anything in the privacy wording that could be clearer.

A few weeks ago I needed to check something in the Action Mailer docs, just a quick lookup. I ended up spending much more time there than expected and found a few features I had no idea existed, even though I've been using Action Mailer in production for a while. One of them lets you see an email before it's ever sent. Another lets you modify an email right before it goes out. And the third one allows you to override the default delivery options dynamically. I figured I probably wasn't the only one who had missed these, so here are three Action Mailer features that caught my attention. If you want to explore more, the official Action Mailer documentation is always a great place to start. 1. Previews Before I found this, testing an email meant sending it to myself, checking my inbox, tweaking the template, and repeating. Turns out ActionMailer has a built-in way to preview emails in the browser, without sending anything. You add a preview class in test/mailers/previews like: class InvitationMailerPreview < ActionMailer :: Preview def team_invitation InvitationMailer . with ( user: User . first , company: Company . first ). team_invitation end end And visit http://localhost:3000/rails/mailers/invitation_mailer/team_invitation . This removes the usual feedback loop of tweaking a template. You just refresh the browser instead. Rails also allows custom preview paths if you want to keep previews in a different location: config . action_mailer . preview_paths << " #{ Rails . root } /lib/mailer_previews" This was a small discovery, but it immediately improved my workflow. 2. Interceptors An interceptor is a hook that runs right before an email is handed off for delivery, letting you modify it. A common use case is preventing mistakes in staging environments. Nobody wants to accidentally send a real looking email from a staging application to an actual customer. Another common approach is redirecting all outgoing mail in staging or development environments to a single defaul

How agents remember - and why deciding what to forget is the real skill An agent that starts every step with a blank mind cannot really pursue a goal. It would reintroduce itself to you on every message, forget what it just tried, and repeat the same mistake forever. Memory is what turns a stateless model into something that accumulates - that knows who you are, what it has already done, and what it learned last Tuesday. This post is about how that works and, more importantly, about the discipline of deciding what an agent should remember at all. The context window is not memory. The first thing to unlearn: a model’s context window is not its memory. The context window is working memory - RAM, not a hard drive. It is finite, it is reset on every request, and every token in it costs money and dilutes the model’s attention. Stuffing an entire conversation history and knowledge base into the prompt does not scale, and past a point it actively hurts - the model loses the important signal in a sea of stale detail. Real memory lives outside the window and is selectively loaded into it when needed. Four kinds of memory Borrowing loosely from cognitive science, agent memory is usually split into four types, and good systems use all of them: Short-term/working memory - the current conversation and the agent’s recent thoughts and observations. Lives in the context window. Long-term episodic memory - a record of what happened : past conversations, decisions, and the outcomes of previous tasks. Long-term semantic memory - facts and knowledge: who the user is, domain information, documents. This is what retrieval-augmented generation pulls from. Procedural memory - how to do things : learned skills, tool-use patterns, and reusable strategies. Short-term memory: the rolling buffer The simplest memory is just keeping recent turns in the prompt. The problem is that conversations outgrow the window, so the standard move is to keep the last few turns verbatim and summarise the older

Four ways to orchestrate business logic in Ruby. One map to find yours. You're building something that involves multiple steps. Charge a card, send an email, update inventory. Simple. Then someone says "What if step 3 fails? What undoes steps 1 and 2?" and suddenly you're evaluating workflow libraries. There are four mainstream approaches in Ruby today — Ruby Reactor , dry-transaction , Trailblazer , and raw Sidekiq jobs. This guide helps you pick the right one — not by ranking them, but by mapping them to the problem you're actually solving. The 30-Second Decision Matrix If you only have 30 seconds, start here: You want... Pick... A simple pipeline — 3-5 steps, top-to-bottom, no parallelism dry-transaction Railway-oriented programming with success/failure tracks, already in the Trailblazer ecosystem Trailblazer A full saga orchestrator — DAG dependency resolution, Sidekiq async, auto-compensation, locks, dashboard Ruby Reactor One-off fire-and-forget jobs, no coordination needed Raw Sidekiq None of these are "better" than the others. They solve different problems. Let's walk through each one. Meet the Libraries dry-transaction (v0.16.0) dry-transaction is a thin, focused gem from the dry-rb ecosystem. It wraps a series of operations in a sequential pipeline with a clean step DSL: class CreateUser < Dry :: Transaction step :validate step :persist step :send_welcome_email def validate ( input ) = # ... def persist ( input ) = # ... def send_welcome_email ( input ) = # ... end Steps run top-to-bottom. If any step returns a Failure , the pipeline stops immediately — no further steps execute. It's a railway under the hood: each step can produce Success(value) or Failure(error) , and the pipeline routes accordingly. What it's great at: Simple, synchronous pipelines. It's the Ruby equivalent of an Either monad chained with bind — clean, predictable, and minimal. If you're already in the dry-rb ecosystem (dry-validation, dry-types, dry-monads), it fits naturally. What it d

1. Introduction Hello from Japan! 🇯🇵 I am an active truck driver in Japan self-studying Python and Web technologies, leveraging my logistics domain knowledge to transition into a Web Engineer. (English is my second language, but I'm excited to share my progress with developers worldwide!) I started my learning journey on May 12, 2026. As of June 4, I have logged 24 days and 86 cumulative hours of study. This article documents my progress, learning roadmap, and project milestones chronologically. Check out my GitHub here👈️ (Note: Most repository documentation and commit messages are currently in Japanese.) 2. The 86-Hour Learning Roadmap A quick breakdown of my 86 hours: 80 hours were dedicated to hands-on development and implementation, and 6 hours were spent on environment configuration, documentation, and workflow optimization. Stages 1–3: Automation & Scraping Projects ① Fundamentals & Web Scraping (25 hours) Learned core Python syntax and built automated scripts to collect targeted web data. ② Puoppo Auto-Analysis System (15 hours) Implemented automated poll data retrieval and AI-powered text analysis on Lubuntu. ③ Bakery Sales Aggregation System (12 hours) Integrated web scraping with automated Excel processing to streamline sales data aggregation. Stage 4: Practical Ruby on Rails ④ rails_practice (23 hours) Explored MVC architecture in Ruby on Rails. Practiced configuration management and Git rebasing to build a solid foundation in web framework operations. Stage 5: Real-World System Integration ⑤ hiroshima-logistics-hub (5 hours) Built a web system to aggregate real-time weather and traffic conditions in the Hiroshima area. Technical Fix : To bypass build crashes on Render (Free Tier / 512MB RAM), I precompiled assets locally and configured SQLite3 database storage in writable directories before deployment. 3. Key Takeaways for Resource-Constrained Development Through these projects, I established three operational principles for developing efficiently within

1. Introduction Hello from Japan! 🇯🇵 I am an active truck driver in Japan self-studying Python, leveraging my logistics domain knowledge to become a Web Engineer. (English is my second language, but I'm excited to share my journey with developers around the world!) I started my self-study journey on May 12, 2026. In this article, I summarize the process of deploying Ruby on Rails 8 to a PaaS (Render Free Tier) and how I tackled the strict resource constraints I ran into. Check out my GitHub here👈️ (Note: Most repository documentation and commits are currently in Japanese.) 2. Environment Development : Lenovo G580 (Lubuntu 24.04 LTS / 16GB RAM / Upgraded SSD) Production : Render (Free Tier: 512MB RAM) Testing Device : Xiaomi 15T 3. Challenges & Solutions ① Git Repository Structure Inconsistency Issue : An unnecessary .git directory existed inside a subdirectory, causing errors during deployment. Solution : Deleted the nested .git directory to restore repository hierarchy integrity. ② Build Failure via Render Free Tier RAM Limit (512MB) Issue : Executing asset compilation on Render triggered Out-Of-Memory (OOM) crashes, forcibly killing the build process. Solution : Precompiled assets locally and committed the static files to the repository, significantly reducing memory usage on the production build server. ③ SQLite3 Write Permission Error Issue : Encountered database write permission errors during CRUD operations in production. Render's file system is read-only by default, except for designated directories (such as storage/ ). Solution : Updated config/database.yml to direct the SQLite3 database file to a path with write permissions (e.g., under storage/ ). 4. Conclusion By applying these workarounds, I successfully verified the deployment and operation of a Rails 8 application on Render's Free Tier. (Please note: Although production runtime works properly, because the setup prioritizes local configurations, some automated CI tests on GitHub currently report errors.

I watched an AI agent send 200 emails to the wrong recipients because I forgot one validation check. The emails were well written. The offers were real. The recipients were just... not our leads. That was early. I learned fast. Every agent I build now has three layers of guardrails before it touches a database or an API. Here's exactly what those layers look like and why they're non-negotiable for production. Input Validation: Your Prompt Is Not a Schema The first mistake people make is trusting the LLM to produce valid output. It won't. Not reliably. I've seen GPT-4 return a JSON key called "emial" instead of "email" in a critical pipeline. One typo, and the whole record is garbage. The fix is a strict validation layer that runs before any data reaches your system. In my AI resume tailor, I use a JSON schema with conditional presence flags. Every field that must be real has a has_* boolean guard. If the LLM tries to fabricate a phone number, the schema rejects it. const resumeSchema = z . object ({ contact : z . object ({ email : z . string (). email (), phone : z . string (). optional (), has_phone : z . boolean () }). refine ( data => { // If phone is present, the guard must be true return data . phone ? data . has_phone : ! data . has_phone }, " Phone number present but has_phone flag is false " ) }) This pattern catches hallucinations before they corrupt your database. The schema is the contract. The LLM is just a suggestion engine. Permission Scoping: Give Agents the Minimum They Need An agent should never have write access to tables it doesn't need. That sounds obvious, but I've seen production systems where a job description rewriting agent had full CRUD access to the user table. When I built the LLM scoring pipeline for a job board platform, I created separate database roles. The scoring agent only had SELECT on the job listings table and INSERT on a scoring results table. It never touched users, applications, or configuration. Even if the prompt was hijack

Users are creating accounts on your new Saas. Yay (and not just family and friends or bots). Yay! Now comes the next step from every marketing handbook: capturing newsletter subscriptions. This article builds on Add Sign Up to Rails 8’ Authentication . Add a simple checkbox to let users opt in to product updates during signup. Store their preference using Rails Vault and manage the subscription with Rails Courrier . First, add Rails Vault and Rails Courrier to your Gemfile: gem "rails_vault" gem "rails_courrier" Rails Vault adds simple and easy settings, preferences and so on to any ActiveRecord model (I recently pushed 1.0.0). Courrier is API-powered email delivery for Ruby apps with support for Mailgun, Postmark, Resend and more. Rails Courrier is the Rails “wrapper” for Courrier. These two gems work really nicely together for this feature. Run bundle install and generate the Rails Vault migration: rails generate rails_vault:install rails db:migrate It creates a new file app/models/user/subscriptions.rb : class User::Subscriptions < Vault vault_attribute :product_emails_subscribed_at , :datetime # Add more subscription types as needed: # vault_attribute :marketing_emails_subscribed_at, :datetime # vault_attribute :weekly_digest_subscribed_at, :datetime end And updates your User model to use this vault: # app/models/user.rb class User < ApplicationRecord + vault :subscriptions has_secure_password has_many :sessions , dependent: :destroy end This keeps subscription data organized without cluttering your User table. More subscription types can be added later without database migrations. Now the plumbing is done, add a checkbox to your signup form in app/views/signups/new.html.erb : <%= form . check_box :product_emails %> <%= form . label :product_emails , "Subscribe to product updates" %> Update the Signup model to accept this parameter: # app/models/signup.rb class Signup include ActiveModel :: Model include ActiveModel :: Attributes attribute :email_address , :stri

LLM Evaluation System Prompts Scored Rubrics Runtime Guardrails: A Practical Guide for Production Learn how to evaluate LLM outputs in production using system prompts, scored rubrics, and runtime guardrails to prevent hallucinations and ensure quality. TL;DR: To evaluate LLM outputs in production, combine system prompts that define evaluation criteria, scored rubrics using LLM-as-a-judge for dimensions like correctness and relevance, and runtime guardrails that filter or flag unsafe outputs. This approach scales better than human review, adapts via prompt changes, and catches failures that status codes miss, as seen in the Air Canada chatbot case. Why Production LLM Evaluation Demands More Than Status Codes A 200 status code only confirms the server processed the request—it says nothing about whether the generated text is factual, safe, or useful. The Air Canada chatbot that invented a non-existent bereavement discount returned perfectly valid HTTP responses, yet the hallucinated policy led to a tribunal ruling against the airline. Production evaluation must therefore separate operational health (latency, error rates) from output quality (correctness, relevance, harmlessness). Consider a typical API call that succeeds operationally but fails qualitatively: import requests response = requests . post ( " https://api.example.com/v1/chat " , json = { " model " : " gpt-4o " , " messages " : [{ " role " : " user " , " content " : " What is Air Canada ' s bereavement policy? " }]}, headers = { " Authorization " : " Bearer $KEY " } ) print ( response . status_code ) # 200 print ( response . json ()[ " choices " ][ 0 ][ " message " ][ " content " ]) # Output: "Air Canada offers full refunds for bereavement-related cancellations..." A 200 status code and a well-formed JSON body mask a completely fabricated policy. To catch this, you need a separate evaluation layer that scores the output against a rubric. LLM-as-a-judge is a common approach, using a second model to assess the

"Context bombing" tricks hacking agents into shutting down before they can do harm.

There's a comfortable thing people say when they see an AI agent query a code map. "Nice crutch. For now." The logic underneath it is reasonable. Coding agents are young. Context windows are small and getting bigger. Models are dumb today and will be smart tomorrow. So a structural index, the thing that hands the agent a dependency graph it would otherwise have to reconstruct, looks like a patch over a temporary weakness. Wait two releases. The model will just hold the whole repo in its head and the map becomes a quaint workaround, like a spellchecker for someone who learned to spell. I build one of those maps Sense . I went looking for the data that would kill it. I didn't find it. I found the opposite. What a map hands an agent is a computed fact. What a better model hands you is a more confident guess . No amount of model progress turns the second into the first, because the difference between them isn't a quality gap that closes with scale. It's a difference of kind. The rest of this piece is the two findings that forced me there. The belief, stated fairly The claim at full strength, because a weak version is easy to knock over. A code map exists to compensate for what the model can't do yet. Today's agent greps, samples, and guesses at structure because it can't read the whole codebase at once. Tomorrow's agent reads all of it, reasons over all of it, and the guessing stops. Bigger windows plus better weights equal no more blind spots. The map is scaffolding you'll tear down once the building stands. If that's true, the right move is to skip the tool and wait. Both findings, in order. Proof one: the best model available was still blind The benchmark ran the same task on thirteen real Ruby repos. Pick the hub model of an app, the Inbox , the MergeRequest , the Spree::Order , and ask the agent to find every place that depends on it before a teardown change. The non-obvious dependents, the ones scattered through concerns and workers and config-string registries, w

You've seen the claim in every Ruby thread for the past year. Ruby and Rails are the most AI-friendly stack. Fewer tokens, less hallucination, the model just writes it cleanly. Half of that claim I'll concede without a fight . The other half I measured, across thirteen real Ruby codebases, and that's where a line shows up, sharp enough to put every repo on one side or the other. Including yours. The half that's true: writing Ruby is solved Start with the part that holds up, because it really does. A model that has seen ten thousand Rails apps knows where the model lives, where the job goes, what a concern does, what has_many implies, before it reads a line of yours. Convention over configuration was always written partly for the next human reading the code. It turns out the model is the next reader too, and the conventions answer half its questions before it asks them. So "write me a service object," "add a scope," "refactor this controller"? The stack carries the model. Fewer wrong guesses, tighter loops, less to hallucinate because the shape is already known. Anyone who builds on Rails has lived this, and the AI-friendly reputation earned it. I'm not here to take that away. I'm here to point out it answers a question nobody dangerous is asking. The half that isn't: navigating Ruby at scale "Can AI write Ruby" is settled. The question that ships broken deploys is different: can AI navigate Ruby? What breaks if this model changes, who depends on it, where the blast radius ends. Reading and navigating feel like the same skill when you're fluent. They are not the same skill for an agent. Reading a file is local, the answer is right there in the text. Navigating is structural, the answer lives in the edges between files, what calls what, what breaks what, and no single file contains it. So I ran the structural question on all thirteen repos. Same task each time: take the hub model, the Inbox , the MergeRequest , the Spree::Order , and find every dependent before a tear