Skip to main content

· 4 min read

AI Features in a Custom Laravel App

AI used to be a separate project bolted onto your app. In a 2026 Laravel build, the core features ship with the framework, switch provider with a setting, and wait for a human before they act.

Anatoly Silko

Anatoly Silko

Founder, Rocking Tech

AI Features in a Custom Laravel App

Three in ten UK businesses now use at least one AI technology, up from fewer than one in ten three years ago (ONS, 2026). For a founder planning a custom app, that changes the brief. Customers and staff increasingly expect the software they use every day to summarise, sort and search for them, and a new build that leaves those features for "phase two" starts life a step behind. The better news is that on a modern framework those features no longer mean a separate integration project, because the building blocks arrive with the foundations.

The five AI features worth planning for

Most useful AI inside a business app falls into five jobs. Summaries turn a long email thread, report or support ticket into a few sentences. Classification sorts incoming items, such as enquiries, invoices or complaints, into the right queue. Semantic search finds records by meaning rather than exact wording, so "late delivery" also surfaces "parcel never arrived".

The other two are chat assistants that answer questions using your own data, and document extraction that pulls names, dates and totals out of PDFs and scans. Adoption is broadest among firms with ten or more staff, where 38% already use AI in some form (ONS, 2026). None of these features needs a research team. Each one needs a clear rule about what the AI is allowed to see and do.

What changed in 2026

Laravel, the PHP framework we build on, released version 1.0 of its official AI SDK on 23 September 2026 (Laravel, 2026). The SDK gives developers one consistent way to call AI models for text, classification, embeddings and agents, and producing a summary is now a single line of code. Version 1.4.0 followed on 9 November 2026 (Packagist, 2026), with agents that recover from a failed step instead of stopping.

Semantic search landed earlier. Laravel 13, released on 17 March 2026, added native vector search, which is the technology behind meaning-based search, directly in the database query layer (Laravel, 2026). It runs on PostgreSQL with the pgvector extension and, since August 2026, on MariaDB 11.7 and later (MariaDB Foundation, 2026). For a founder, the practical point is simple: smart search no longer requires a separate search service with its own bill.

Laravel 13 also receives security fixes until 17 March 2028 (Laravel, 2026). An app started on it today launches on a supported foundation, with a routine upgrade path after that.

Why provider independence matters

AI models are retired on a published schedule. OpenAI announced on 22 April 2026 that gpt-3.5-turbo would shut down, and it did on 23 October 2026 (OpenAI, 2026). Anthropic retired Claude Sonnet 4 on 15 June 2026 and commits to at least 60 days' notice before retiring a public model (Anthropic, 2026).

Retirements are a normal part of how AI providers improve their products, and they are easy to absorb when the app is built for them. The Laravel AI SDK supports OpenAI, Anthropic, Gemini, Azure, Bedrock, Mistral, Groq, xAI, DeepSeek, Cohere, OpenRouter, Ollama and any OpenAI-compatible service (Laravel, 2026). Developers can also list several providers so the app fails over automatically if one is unavailable.

The same design keeps commercial options open. If a different model becomes cheaper or better at your particular task, switching is a decision about quality and price, with no engineering project attached.

Keeping a person in charge

The most valuable AI features are the ones that take action: updating a record, sending a reply, issuing a refund. Those are also the ones that need a firm boundary. The National Cyber Security Centre advises deciding, before an agent runs, when it should stop and seek human approval, and making sure those approvals cannot be skipped (NCSC, 2026).

The Laravel AI SDK builds that boundary in. Developers can mark any tool as requiring approval, and the agent pauses at that point (Laravel, 2026). A member of staff then approves the action, rejects it with a reason, or edits it before it goes ahead.

The regulator is moving in the same direction. The ICO consulted on new guidance on automated decision-making between 31 March and 29 May 2026, following the Data (Use and Access) Act 2025 (ICO, 2026). Its draft makes clear that human oversight has to be meaningful, so a reviewer needs real authority to challenge or override the AI. An approval step designed in from the first sprint is far easier to evidence than one retrofitted after launch.

Faster, steadier delivery behind the scenes

The same ecosystem has made the engineering itself quicker to check. Pest 5, the testing framework many Laravel teams use, now re-runs only the tests affected by a change. Laravel's founder reported that it took the Laravel Cloud test suite of more than 19,000 tests from three minutes to five seconds (Otwell, 2026). Fast tests mean every change, including AI-assisted code, is checked before it reaches your users.

Coding with AI agents works best on a well-established framework, where the rules are already set.

Hosting has followed the same path. Laravel Cloud's newer compute can scale to zero when nobody is using the app and wake in under 500 milliseconds (Laravel Cloud, 2026). For an early product with uneven traffic, that keeps running costs tied to real use.

Planning AI into your build

The cheapest time to decide on AI features is before any code is written. Which records should the AI read, which actions need a human sign-off, which database supports the search you want, and which provider suits your data: each answer shapes the architecture.

At Rocking Tech, AI-enhanced apps are one of the things we build on Laravel and Vue: intelligent automation, document processing and smart recommendations, with AI used where it adds value rather than as a gimmick. Every build starts with a three-week Discovery Sprint at £4,950 +VAT, which produces the architecture, a costed phased roadmap and a fixed price. The fee is credited in full against the build if you sign within 60 days, and we're also a Laravel Partner, with a direct line to the team behind the framework.

If you are still weighing whether a custom build is the right call at all, our guide on when to build custom vs stick with SaaS is a good place to start, and what custom MVP development costs in the UK covers the budget side. The AI layer is now the easy part to plan, provided it is planned.

Every client review on Trustpilot: 5 stars →

Ready to build what nothing off-the-shelf can do?

Book a free 30-minute call. We’ll discuss what you’ve tried, what’s not working, and whether a Discovery Sprint is the right next step. If it is, you’ll know exactly what to build – and the £4,950 comes off the build if you go ahead.