Laravel 13 arrived on March 17, 2026, with first-party AI tooling among its changes.1 For businesses running custom Laravel applications, that creates useful possibilities for adding features such as text extraction, classification and assisted search.
The most promising starting point is a specific job inside an existing workflow.
Perhaps staff receive unstructured service requests and spend time identifying the location, type of work and missing information. AI could help prepare a structured draft. The application can then apply its normal rules before anyone treats that draft as an accepted job.
That is a more useful plan than adding a chat box and waiting for the business case to wander over.
Give AI a defined place in the application
Laravel’s AI SDK was introduced in beta in February, with a unified interface for working with multiple providers.2 Laravel 13’s release notes bring that first-party AI direction into the framework’s broader story.
A consistent programming interface can help keep provider integration organized. It does not make every provider’s capabilities or responses interchangeable. The selected model, supported features and behavior still need checking for the intended task.
In the hypothetical service-request workflow, I would define a small output: proposed location, requested work category, relevant details and questions that remain unanswered. Staff should be able to compare that draft with the original request.
Structured output can make the result easier for software to parse. A response fitting a schema can still contain an incorrect location or an unsupported assumption. The application needs to validate the business meaning as well as the format.
For instance, a category can be valid in the database while being wrong for this particular request. A clear review step lets staff correct it before it affects scheduling.
Curious what an AI agent could do for your team? See our AI agent services.
Keep decisions and permissions in the right place
Existing domain rules should continue to govern accepted work. Service-area checks, customer permissions, pricing and availability belong in controlled application logic.
If an AI feature can call tools, expose specific operations with appropriate authorization. A tool that retrieves one customer’s permitted records is a different proposition from a broad query that happens to receive a polite instruction about privacy.
Treat incoming material as information to inspect. An instruction buried in a customer message should not gain the authority to change the application’s rules. The surrounding code and tool permissions need to maintain that boundary.
Operational behavior deserves a similarly ordinary plan. What should staff see when the provider is unavailable, the result is incomplete or the request needs manual attention? A useful feature can fall back to the established process without leaving the work stranded.
Related read: Laravel 11 got leaner. Your business rules still need a home.
Test representative requests, including ambiguous ones, before treating the output as dependable. If the team changes providers or models, repeat the relevant checks. A backup provider might return a perfectly formed response with different judgments.
The framework upgrade is also a separate maintenance decision. Laravel 13 requires at least PHP 8.3, so hosting and package compatibility need review alongside any proposed AI feature. A new feature and a runtime change do not have to become one enormous parcel.
I like the direction: AI capabilities can sit closer to the application and its existing tools. The business still benefits from a modest beginning, clear review and a result that staff can understand.
Let the AI help unpack the request. It does not need to appoint itself operations director while everyone is making coffee.