AI coding tools are now part of everyday development. Copilot, Cursor, Claude Code, Windsurf. If you're building with October CMS and you're not using at least one of these, you're leaving productivity on the table.

But using them well takes more than installing an extension. There's a growing body of knowledge around how to structure your work so AI agents produce good code instead of slop. Matt Pocock's AI Hero has been leading this conversation, and his core insight is simple: your codebase and your workflow matter more than which model you use.

The good news for October CMS developers is that you're already working in an architecture that AI tools navigate well, and we've built tooling (October CMS Boost) to make it even better. This post covers the key concepts from the AI-assisted development space and shows you how to apply them to your October CMS projects today.

Concept 1: Deep Modules

The most important idea in AI Hero's framework comes from John Ousterhout's A Philosophy of Software Design: deep modules.

A deep module has a simple interface controlling a large implementation. The AI reads the interface, understands what the module does, and delegates the internals. A shallow module does the opposite: tiny implementation behind a complex web of imports and dependencies that the AI has to trace through.

"Bad codebases have lots of shallow modules with big interfaces. Good codebases have a few big modules with simple interfaces."

October CMS plugins are deep modules by default. Each plugin has a single entry point (Plugin.php) that declares everything the module provides through explicit methods: registerComponents(), registerNavigation(), registerPermissions(), registerSettings(). An AI agent reads this one file and knows the full surface area of the plugin without opening a single other file.

How to lean into this:

• Keep your Plugin.php as the source of truth. If you're registering things in service providers or boot methods that could go in the declarative registration methods, move them.
• When you ask an AI to extend your plugin, point it at Plugin.php first. That gives it the map.
• Resist the urge to break your plugin into many tiny classes just for "cleanliness." Fewer modules with clear interfaces is better for both humans and AI.

Concept 2: Declarative Metadata

AI parses data structures faster than it traces execution paths. When your configuration is expressed as YAML or typed arrays instead of procedural code, the AI reads it once and understands the full picture.

October CMS is built on this pattern. Tailor blueprints, fields.yaml, columns.yaml, config/ files, component property declarations. All of these express behavior as data. When an AI tool reads a Tailor blueprint like this:

handle: Blog\Post
type: stream
name: Post
drafts: true

groups:
    regular_post:
        name: Regular Post
        fields:
            content:
                label: Content
                type: richeditor
            banner:
                label: Banner
                type: fileupload
                mode: image
            categories:
                label: Categories
                type: entries
                source: Blog\Category

It immediately understands the data model, the field types, and the relationships. It can generate new blueprints, add fields, or modify the structure by pattern-matching. No ORM knowledge required, no migration files to trace.

The same applies to fields.yaml in your plugin models. The AI reads the YAML and knows every form field, its type, its validation, its conditional visibility. It can add a new field by copying the pattern of an existing one.

How to lean into this:

• Tailor blueprints are ideal for content schemas (blogs, FAQs, team pages) where you don't need custom model logic. For anything with real functionality, plugins give you full control over the models and database.
• Keep your fields.yaml and columns.yaml files well-organized. AI tools use them as a reference for how your backend forms and lists work.
• When asking an AI to add a backend form field, show it the existing fields.yaml first. It will match the pattern exactly.

Concept 3: Tracer Bullets

AI Hero's tracer bullet approach is about building small, end-to-end slices of functionality rather than large horizontal layers. The idea comes from The Pragmatic Programmer and it's especially important when working with AI.

AI's natural tendency is to produce a complete solution all at once. Ask it to "build a product review system" and it will generate models, controllers, components, partials, and migration files in one shot. The result is usually a large block of code with subtle integration errors that take longer to fix than it would have taken to build incrementally.

The fix is to think in vertical slices. Instead of "build a review system," break it down:

1. First: create the model and migration. Test it.
2. Then: create the component with just onSubmitReview(). Test it.
3. Then: create the default partial with the form markup. Test it.
4. Then: add the display logic. Test it.

Each slice goes through all layers (model, component, partial) but only does one thing. You validate as you go, and each step gives the AI a working foundation to build on.

How to apply this in October CMS:

• When building a new component, start with componentDetails(), defineProperties(), and one AJAX handler. Get it working. Then add the next handler.
• When building a Tailor blueprint, start with two or three fields and a basic page template. Validate the rendering. Then expand.
• Use the php artisan tinker or the backend to verify each step before moving to the next. The AI performs better when it builds on verified code rather than assumptions.

Concept 4: Test-Driven Development for AI

TDD is the most consistent way to improve AI agent output. Not because of code coverage metrics, but because tests create a feedback loop that AI needs to self-correct.

Without tests, the AI generates code and has no way to know if it works. With tests, the AI can run them, see failures, and iterate. This is the "red-green-refactor" cycle applied to AI-assisted development:

1. Write a test that describes the desired behavior
2. Ask the AI to write the implementation
3. Run the test
4. If it fails, the AI can read the failure and fix it
5. Refactor once it passes

October CMS has a testing infrastructure built on Laravel's test framework. You can write feature tests for your plugin's components, unit tests for your models, and integration tests for your AJAX handlers.

How to apply this:

• Before asking an AI to build a feature, write the test first (or ask the AI to write the test). Then ask it to make the test pass.
• For component AJAX handlers, test them by simulating POST requests. The AI can read the test to understand what the handler should accept and return.
• For Tailor blueprints, validate by checking that the expected fields exist and render correctly.
• If you're not writing tests today, start small. Even a few tests on critical paths give the AI something to work with.

Concept 5: The Grill-Me Pattern

Before you start building, let the AI interview you. AI Hero calls this the /grill-me skill: ask the AI to "interview me relentlessly about every aspect of this plan until we reach a shared understanding."

This is especially useful for October CMS projects because the platform has specific patterns that the AI may not know about. By front-loading a conversation, you teach the AI about your project's structure before it starts generating code.

Example prompt:

I'm building a plugin for October CMS 4.x that handles event bookings. Before we start coding, I want you to interview me about the requirements. Ask me about the data model, the frontend components, the backend controllers, the AJAX handlers, and any events I need to fire. Walk through each decision one at a time.

A good AI will ask you 15-30 clarifying questions. By the end, it has a detailed understanding of what you need, expressed in October CMS terms (components, handlers, blueprints, partials). The code it generates after this process is dramatically better than a cold start.

Concept 6: Giving AI Context About Your Project

AI Hero's guide to AGENTS.md explains how to create a project-level configuration file that tells AI tools how to work with your codebase. Claude Code uses CLAUDE.md, Cursor uses .cursorrules, but the concept is the same: give the AI a brief orientation before it starts reading code.

For October CMS projects, we've built a tool that handles this automatically: October CMS Boost.

October CMS Boost

Boost is a development package that extends Laravel Boost with October CMS-specific capabilities. Install it and your AI agent immediately understands October CMS conventions without you having to explain them.

composer require october/boost --dev
php artisan boost:install

It provides three layers of AI context:

Guidelines are loaded automatically into every AI conversation. They cover the critical differences from vanilla Laravel (don't suggest Livewire, Inertia, or Blade), the plugin architecture, model patterns, the event system, and naming standards. This eliminates the most common AI mistake with October CMS: generating standard Laravel code instead of idiomatic October CMS code.

Skills are on-demand knowledge modules that activate when the AI needs them. There are six, covering plugin development, Tailor blueprints, backend controllers, theme development, the AJAX framework, and model patterns. When you ask the AI to build a component, it loads the theme development skill. When you ask it to create a blueprint, it loads the Tailor skill.

MCP Tools give the AI live access to your project. It can search the October CMS documentation, inspect your Tailor blueprints, read your plugin registrations, and examine your theme structure. Instead of guessing, the AI can look at your actual project and work with what's there.

This is the AGENTS.md concept taken further. Rather than writing a static markdown file, Boost provides dynamic, context-aware guidance that adapts to what the AI is doing.

Writing Your Own AGENTS.md

Even with Boost installed, you can add a project-specific CLAUDE.md or AGENTS.md for things Boost doesn't know about. Keep it minimal:

• One-line project description: "This is an October CMS 4.x site with Meloncart eCommerce and a custom events plugin."
• Project-specific conventions: Anything unique to your project that differs from October CMS defaults.
• Testing: "Run tests with php artisan test Acme.MyPlugin or php artisan test Acme.MyPlugin --filter=MyTestClass."

The AI Hero guide recommends ruthless minimalism: only include what the AI can't discover by reading the code. Between October CMS's strong conventions and Boost's automated context, you shouldn't need much.

Putting It All Together

Here's a practical workflow for building a feature in October CMS with AI:

1. Grill-me first. Describe what you want to build and let the AI interview you about the specifics. Answer in October CMS terms (components, partials, blueprints).

2. Write the test. Or ask the AI to write a test for the behavior you want. This anchors the implementation.

3. Tracer bullet. Build the smallest end-to-end slice first. One component, one handler, one partial. Verify it works.

4. Iterate. Add the next slice. The AI builds on verified code, not assumptions.

5. Refactor. Once the feature works, ask the AI to consolidate. "Can this be simplified? Are there deep module opportunities?"

The concepts here aren't October CMS-specific. They come from the broader AI-assisted development community and they work with any framework. But October CMS's declarative, convention-driven architecture gives you a head start, and Boost closes the remaining gap by teaching AI agents the conventions you already know.

The patterns AI agents need (deep modules, metadata-driven config, predictable structure) are already built into the platform. You just need to meet the AI halfway with the right workflow.

For a deeper look at how these principles map to specific architectural patterns, check out Matt Pocock's AI Hero, particularly How to Make Codebases AI Agents Love and 5 Agent Skills I Use Every Day. And if you haven't already, install October CMS Boost to give your AI tools native October CMS understanding out of the box.