Laravel just shipped a first-party AI SDK. Here's what it actually means for your team.

Last week, the Laravel team released something that's been brewing since Laracon India: a first-party AI SDK. It's now officially out as v0.1.2, installable via composer require laravel/ai, and it does exactly what you'd expect from a Laravel package: it wraps multiple AI providers into a clean, opinionated API so you don't have to juggle six different SDKs and their documentation.

The SDK supports OpenAI, Anthropic, Google Gemini, Groq, and xAI out of the box. You get a new Agent pattern (scaffolded via php artisan make:agent), built-in conversation persistence, streaming responses, image and audio generation, and even vector search for RAG workflows if you're running PostgreSQL with pgvector. It builds on the community Prism package under the hood, adding Laravel-specific tooling on top.

Our take

We've been building AI-powered features into client projects for the past year, mostly using direct API calls, custom service classes, and a fair amount of duct tape. The honest truth? Every project ended up with its own slightly different abstraction for talking to OpenAI or Anthropic. Different error handling, different retry logic, different ways of storing conversation history. It worked, but it wasn't elegant, and it definitely wasn't reusable across projects.

So a first-party SDK that standardizes this? That's genuinely useful. The Agent class pattern in particular is the right idea. When we build cloud-native apps for enterprise clients, we already push teams toward dedicated service classes for complex operations. Having an official make:agent command with a defined structure for instructions, tools, and output schemas means less bikeshedding about where AI logic lives in the codebase.

The provider failover feature is also worth calling out. You can pass an array of providers like ['anthropic', 'openai'] and the SDK handles fallback automatically. From our PHP and Docker work on high-availability systems, we know that provider outages are not hypothetical. Rate limits get hit. APIs go down at 2am. Having automatic failover without writing custom retry logic is a genuine time-saver.

Where we'd pump the brakes

It's v0.1.2. That number should give you pause before you wire this into anything mission-critical.

The vector search features require PostgreSQL with pgvector, which isn't what most Laravel shops are running. If you're on MySQL (and statistically, you probably are), the RAG capabilities won't be available to you without a database migration. That's not a small ask for existing production apps.

There's also a question that the SDK itself can't answer: how you should handle AI costs. The SDK makes it trivially easy to call AI providers, which means it's also trivially easy to accidentally burn through API credits. We've seen this pattern in client projects already, where a developer adds an AI call inside a loop or a queue job, and suddenly the monthly bill jumps by a few thousand. The SDK has a nice UseCheapestModel attribute, but cost governance is still your problem to solve.

And this is a beta. The API surface will change. If you're building a product that ships to paying customers, you need to be comfortable with that reality.

What about the broader Laravel ecosystem?

The AI SDK isn't landing in isolation. It joins Laravel MCP (Model Context Protocol, released back in September) and Laravel Boost (the AI-assisted development package). Together, they signal where the Laravel ecosystem is heading: AI isn't an add-on anymore, it's becoming a first-class part of the framework.

Laravel Wayfinder, which hit beta in January, is also worth watching. It syncs PHP and TypeScript types automatically, generating routes and validation from your Laravel app. If you're running a Laravel backend with a React or Vue frontend (and many of our development projects do exactly that), this could save real hours on keeping your API contracts in sync.

A quick note on iOS in Japan

On the mobile side, something significant happened in December that's still rippling through: Japan's Mobile Software Competition Act (MSCA) went into effect, and Apple has now opened iOS to alternative app marketplaces and third-party payment processing in Japan. This makes Japan the second market after the EU where developers can distribute iOS apps outside the App Store.

In our native iOS and Android projects, especially for clients targeting international markets, this kind of regulatory shift matters. The fee structures range from 5% to 26% depending on your distribution method and payment choice in Japan. For apps with heavy digital goods sales (gaming, subscriptions, content), the math on alternative distribution is starting to make sense in specific markets.

But here's the catch that keeps coming up during our penetration testing and security review engagements: apps distributed through alternative marketplaces only go through Apple's "Notarization" process, which is a lighter review than full App Review. If you're building in healthcare, finance, or any regulated space, you need to think carefully about whether the fee savings are worth the reduced trust signal that comes with App Store distribution. Our experience with Swiss enterprise clients tells us that most regulated businesses will stick with the App Store for now.

What to do this week

If you're a Laravel team:

1. Install the AI SDK in a throwaway project and build something small. The agent pattern is worth understanding even if you don't ship anything with it yet.
2. If you're already calling AI APIs directly, sketch out what a migration path to the SDK would look like. Don't migrate yet, but know the shape of it.
3. Audit your current AI spend. Before making it easier to call AI providers, know what you're paying today.

If you're shipping iOS apps internationally:

1. Read Apple's MSCA compliance page. Even if you don't have Japanese users today, the pattern of regulatory pressure is clear, and more markets will follow.
2. All Apple Developer Program members need to accept updated license terms by March 17, 2026. Don't let that deadline sneak up on you.

We're going to be integrating the Laravel AI SDK into a couple of internal projects over the next few weeks and will share what we learn. If you're wrestling with AI integration in a Laravel codebase, or trying to figure out what Japan's App Store changes mean for your mobile distribution strategy, let's talk.

You can read more about how we approach these problems across our services or dig into past work for context on the kinds of projects we're talking about.