Everyone is rewriting everything. I added one Composer package and kept shipping.
Brittany Chiang wrote a piece called “Welcome to the AI Parade.” Her thesis is simple. Software engineering has changed and there is no going back.
She is right. But I think the parade looks different depending on where you are standing.
If you work in a React/TypeScript startup in San Francisco, the parade probably looks like a full rewrite. New agent framework every quarter. New build tool every month. A constant feeling that whatever you shipped last week is already legacy.
If you run a Laravel monolith that serves real clients in Southeast Asia, the parade looks more like a slow upgrade. Same codebase. Same deployment pipeline. A few new tools bolted on. Business as usual, except faster.
I joined the AI parade. I did not tear down what was already working to do it.
What the Parade Looks Like From Here
My stack has not changed in three years. Laravel 12. PHP 8.4. MySQL. Redis. Deployed on Forge. No microservices. No Kubernetes. No event-driven architecture. Just a monolith that handles RM 2M+ in contracts annually across 11 active projects.
When the AI coding tools started appearing, I watched from the sidelines. Copilot in 2022 felt like fancy autocomplete. ChatGPT in late 2022 was impressive but unreliable for production PHP. By mid-2024, Cursor was getting interesting. By 2025, the tools were good enough to integrate into real workflows.
But here is the thing. I did not need to rewrite my stack to use them. I needed to give them context about my stack.
That is the difference between the AI parade for greenfield projects and the AI parade for existing systems. New projects can adopt AI-first architectures. Existing projects need AI to understand what already exists.
Laravel Boost Changed the Game
In August 2025, Taylor Otwell previewed Laravel Boost at Laracon US. Two weeks later it shipped as a public beta. I installed it the same day.
composer require laravel/boost --dev
php artisan boost:installTwo commands. That was the entire “AI transformation” of my codebase.
Boost is an MCP server. It runs inside your Laravel project and gives AI agents deep access to your application’s structure. Your database schema. Your routes. Your Eloquent models. Your installed packages and their versions. Your configuration values. Even your log files.
Before Boost, asking Claude Code to add a feature to my invoicing module was a coin flip. Sometimes it would generate code using a package I did not have installed. Sometimes it would create a migration with column types that did not match my existing schema. Sometimes it would ignore my naming conventions entirely.
After Boost, the AI agent knows my application. It can run Tinker to test code. It can query my database to understand the data. It can search version-specific Laravel documentation. Boost even ships with composable AI guidelines that teach the agent Laravel conventions specific to my installed packages.
The difference is not subtle. It is the difference between hiring a contractor who has never seen your codebase and hiring one who has spent a week reading it first.
My Actual Workflow Now
Here is what a typical feature implementation looks like in February 2026.
Step 1: I write the spec. Not in code. In plain English. “Add a payment reminder system. Tenants should be able to configure reminder intervals. Send reminders via email 7, 14 and 30 days after invoice due date. Track which reminders have been sent.”
Step 2: Claude Code with Boost generates the first draft. Migration, model, controller, Form Request, Mailable, scheduled command. Because Boost gives it access to my schema, it knows my invoices table uses total_cents instead of total. It knows I have a BelongsToTenant trait. It follows my naming patterns because the guidelines tell it to.
Step 3: I review. This is the step that matters. The AI gets 80% right. The migration is clean. The basic CRUD works. But the edge cases are always wrong. What happens when a tenant has 10,000 overdue invoices? The scheduled command needs chunking. What about invoices that are partially paid? The reminder logic needs to check paid_cents against total_cents, not just the status column.
Step 4: I fix the 20%. This takes about the same time the AI saved me on the 80%. But the cognitive load is different. I am reviewing and correcting, not writing boilerplate from scratch. My brain is in quality mode, not production mode.
Step 5: I write tests. Sometimes the AI writes them. Sometimes I do. But I always run them and I always review them.
Total time for this feature: about 4 hours. Before AI tools, the same feature would take a full day. Not because the code was harder, but because the boilerplate was tedious.
What I Did Not Do
I did not rewrite my monolith as a set of microservices so that AI agents could work on isolated components. I did not switch to TypeScript because “AI understands it better.” I did not adopt a new framework because some influencer said it was “more AI-native.”
I installed a Composer package. I configured my IDE. I kept shipping.
The monolith is still a monolith. The deployment is still forge deploy. The database is still MySQL. PHP is still the language that powers 77% of websites on the internet, according to W3Techs.
The AI parade did not require me to change my foundation. It required me to open the door and let the new tools in.
Why Laravel Monoliths Are Secretly Perfect for AI
There is a reason Boost works so well with Laravel. And it is the same reason Laravel has always been productive: convention over configuration.
When you ask an AI agent to add a controller, it knows where to put it. app/Http/Controllers. When you need a migration, the file naming and location are predictable. When you ask for a Form Request, the agent knows the pattern because every Laravel developer uses the same pattern.
Compare this to a custom Node.js project where every team has a different folder structure, different ORM, different validation library and different routing approach. The AI has to guess. With Laravel, it does not.
This is the quiet advantage that nobody talks about in the AI parade discourse. Opinionated frameworks with strong conventions are easier for AI to work with. Not because the AI is smarter, but because the codebase is more predictable.
Laravel’s structure is basically a prompt.
The Real Cost of Joining the Parade
Here is what I actually spent to “join the AI parade” for my Laravel practice:
Claude Max subscription: $200/month. This gives me Claude Code with Opus 4.6, which handles complex multi-file Laravel changes without running out of context.
Laravel Boost: Free. Open source. MIT license.
Time to set up: About 30 minutes per project.
Changes to my stack: Zero.
Changes to my deployment pipeline: Zero.
Changes to my hiring criteria: One addition. I now look for developers who can review AI-generated code effectively, not just write code from scratch.
Compare this to the companies rewriting their entire architecture to be “AI-native.” Some of them will come out ahead. Many of them will spend six months rebuilding what they already had, plus AI.
I spent 30 minutes and a Composer install.
The Parade Has Room for Everyone
The AI conversation in tech is dominated by two camps. The first camp says AI changes everything and you need to start over. The second camp says AI is overhyped and you should ignore it.
Both camps are wrong.
AI is real. The productivity gains are measurable. My team ships features faster. Our proposal turnaround dropped from 7 days to 3. Our code review process is quicker because the AI catches the obvious stuff before a human looks at it.
But “real” does not mean “revolutionary for everyone.” For greenfield projects, AI might justify a completely new architecture. For a profitable Laravel monolith with paying clients and a proven deployment pipeline, AI is an accelerator. Not a replacement.
I joined the parade. My monolith is still marching. And honestly, it is marching faster than before.