Laravel AI SDK and Xcode 26.3 Agentic Coding

Two big releases landed this week that are worth paying attention to, whether you're a PHP shop, a mobile team, or both.

On February 5th, Laravel announced its official AI SDK, a first-party package that gives Laravel apps a unified way to talk to AI providers like OpenAI, Anthropic, and Google Gemini. You install it with composer require laravel/ai, scaffold Agent classes with Artisan, and get built-in support for text generation, image creation, audio, streaming, structured output, and even vector search with PostgreSQL's pgvector. It's currently at v0.1.2, built on top of Prism under the hood, and Taylor Otwell is doing a live walkthrough on February 9th.

Days earlier, Apple released the Xcode 26.3 Release Candidate, and this one is not a small update. Xcode now supports full agentic coding, meaning AI agents from Anthropic (Claude Agent) and OpenAI (Codex) can autonomously write code, build your project, run tests, capture SwiftUI Preview screenshots, and iterate on errors without a developer copying compiler output back into a chat window. Apple also adopted the Model Context Protocol (MCP) as an open standard, so any MCP-compatible agent can plug into Xcode. That last part is genuinely surprising, coming from Apple.

Both of these stories are connected by the same thread: AI is being woven directly into the tools we use every day, not as a bolt-on, but as first-party infrastructure.

The Laravel AI SDK: a sensible move, with a caveat

We've been building PHP applications with Laravel for years, and I keep coming back to the same pattern: every time the framework absorbs something that used to require a third-party package, it gets better. Queues, broadcasting, caching, authentication. Laravel's strength has always been that the boring plumbing works out of the box.

AI integration was heading that way. Before this SDK, if you wanted to call OpenAI from a Laravel app, you'd pull in openai-php/client, write your own service classes, manage conversation state manually, and handle provider-specific quirks yourself. If you needed to switch from OpenAI to Anthropic mid-project (which happens more than you'd think, especially when a client's compliance team suddenly has opinions about data residency), you were looking at a partial rewrite.

The SDK fixes the provider-switching problem. The Agent pattern is clean. The Artisan scaffolding (php artisan make:agent) feels like Laravel. And the fact that vector store features require PostgreSQL with pgvector is a smart constraint, not a limitation. From our PHP and Docker work with enterprise clients, PostgreSQL is already the default database choice for anything that needs to scale or comply with European data requirements.

The caveat: it's v0.1.2. It's a beta. If you're running a production SaaS, don't rip out your existing AI integration layer this weekend. Start a branch, try it on a new feature, and see how it feels. For greenfield projects, though, this is probably the right starting point now.

One thing the community hasn't talked about enough yet: the MCP integration. Laravel now has first-party packages for AI SDK, Boost (for AI-assisted development), and MCP (for exposing your app's functionality to AI clients like ChatGPT or Claude). Those three together mean your Laravel application can both consume AI services and be consumed by AI agents. That's a real architectural shift, and teams building APIs should be thinking about it now, not later.

Xcode 26.3: impressive demo, real questions

In our native iOS and Android projects, we've watched AI coding assistants go from novelty to daily tool over the past two years. But Xcode 26.3 is doing something different from autocomplete and chat suggestions.

The agents in Xcode 26.3 can discover your project structure, consult Apple's developer documentation, write code across multiple files, build the project, run tests, capture screenshots of the running UI via Xcode Previews, and then visually inspect those screenshots to check if the implementation matches the request. When the agent hallucinates and writes code that doesn't compile, it can see the errors and fix them itself. Apple also creates automatic rollback checkpoints as the agent works, so you can revert at any point.

During Apple's press event, a developer described previous attempts to use AI agents with Xcode as "horrible." Apple acknowledged this. The argument is that earlier tools were blind. They could only see what you pasted into them. Now the agents have real IDE access.

Here's what I'm actually thinking about, though. We do penetration testing for gaming studios. We review code for security flaws. The question that keeps nagging at me is: what happens when an autonomous agent adds an entitlement to your project because it thinks it needs access to a protected API? Apple mentioned agents can do this. That's convenient when you're prototyping. It's a security concern when nobody's reviewing the diff.

MCP adoption is the more interesting angle here. Apple going with an open standard instead of building something proprietary is unusual. It means any MCP-compatible tool, not just Claude and Codex, can integrate with Xcode. If you're building developer tools or security scanners, this opens a real opportunity. And if you're an iOS team that has been using Claude Code from the terminal, you can now connect it to Xcode via MCP and get Preview captures without leaving the CLI.

What development teams should do this week

Here's what I'd actually recommend:

If you're a Laravel team, install the AI SDK on a test project. Read the docs at laravel.com/ai. Pay particular attention to how Agents handle conversation persistence and failover between providers. If you're building anything that touches AI, this should be your default integration layer for new work. Don't migrate production yet.

If you're an iOS team, download Xcode 26.3 RC and try the agentic coding on a non-critical feature branch. Start with something small, like generating a new SwiftUI view with form validation. See how the agent handles your project's specific architecture before you trust it with anything complex. And review every diff. The rollback checkpoints are nice, but they're not a substitute for code review.

If you're a team that does both web and mobile (like us), the MCP thread connecting these two announcements is the thing to watch. Laravel can now expose app functionality via MCP. Xcode can now consume tools via MCP. The protocol is becoming the standard interface between AI and everything else. Understanding MCP now will pay off over the next 12 months.

And regardless of your stack: talk to your security team about AI agent access patterns. When configuring Cloudflare or reviewing cloud infrastructure, we've seen how quickly automated systems can introduce unintended exposure. AI agents that can modify project settings, add entitlements, and interact with APIs are a new surface area. Treat them like you'd treat any other automated system with elevated privileges.

We've seen this pattern before in our work with enterprise clients: new tooling arrives, teams rush to adopt, and the security review happens six months later when something breaks. Don't be that team.

If your team is figuring out how to integrate AI into your Laravel stack, or you're evaluating how agentic coding fits into your iOS development workflow without introducing new risks, let's talk. We've been building across both worlds for a while now, and these questions are the exact ones we're helping teams answer right now.