From AI Replacement Fear to AI Command: A PHP Developer's Laravel AI SDK Journey

The Wake-Up Call

An intern suggested using Cursor for an order export feature. The author tried it: the same requirement took four minutes to produce clean, working code versus half a day of manual work. This triggered a hard look at market data.

Market Reality

Malt Tech Trends 2026 shows PHP skill demand down 33% year-over-year (Java -32%, C++ -25%), while AI engineer and data scientist roles surge. The message: companies want people who can build with AI, not just write CRUD. Meanwhile, GitHub Copilot, Cursor, and Claude Code already generate complete PHP auth modules, CRUD APIs, and payment callbacks in seconds.

Turning Point: Laravel AI SDK

While reading Laravel docs, the author discovered Laravel AI SDK (released February 2026), a unified SDK for PHP developers to build AI-powered applications across 14 providers.

First Experiment: A Customer Service Agent

Goal: an agent that reads product data from the database and answers price, stock, and spec questions. Using Laravel AI SDK, the core Agent class is minimal:

<?php

namespace App\Ai\Agents;

use Laravel\Ai\Contracts\Agent;
use Laravel\Ai\Promptable;

class ProductSupportAgent implements Agent
{
    use Promptable;

    public function instructions(): string
    {
        return '你是一个产品客服助手。根据提供的产品数据回答用户问题。如果数据库中找不到相关信息，如实告知。';
    }
}

Controller invocation:

$agent = new ProductSupportAgent();
$answer = $agent->prompt("这个商品现在有货吗？");

The agent runs directly on Laravel's ecosystem — Eloquent for queries, queues for async, existing auth and middleware. No separate Python microservice, no second codebase. Deployed to operations, it handled ~20,000 consultations over six months with zero crashes.

Second Experiment: Giving the Agent "Hands" via Tools

Chat-only agents are just glorified auto-responders. The author added a CheckStock tool so the agent can query live inventory:

<?php

namespace App\Ai\Tools;

use Laravel\Ai\Contracts\Tool;

class CheckStock implements Tool
{
    public function name(): string
    {
        return 'check_stock';
    }

    public function description(): string
    {
        return '根据商品ID查询实时库存数量';
    }

    public function parameters(): array
    {
        return [
            'product_id' => ['type' => 'integer', 'required' => true],
        ];
    }

    public function handle(array $params): array
    {
        $product = \App\Models\Product::find($params['product_id']);

        return [
            'name' => $product?->name,
            'stock' => $product?->stock ?? 0,
            'available' => ($product?->stock ?? 0) > 0,
        ];
    }
}

Register the tool in the Agent's tools() method. When a user asks "Does product A have stock?", the AI decides to call the tool, fetches real-time data, and answers. No manual routing logic, no if-else chains. The realization: the developer isn't being replaced; they're directing AI.

Third Experiment: Letting AI Write the AI Code

Manually crafting every Agent and Tool becomes its own grunt work. The author used Claude Code's Plan Mode :

claude "Plan the implementation of a multi-channel notification system for this Laravel app. Use Laravel AI SDK to create an agent that decides which channel to use."

Claude Code produces a full plan — classes, responsibilities, Agent instructions, required Tools. After approval, it writes the code, runs PHPUnit, and reports failures. The developer's role shifts from "code writer" to "code reviewer". Coding speed increases ~10x, but judgment and experience remain essential. Claude Code's MCP integration also exposes local databases, logs, and runtime profilers to the AI, accelerating root-cause analysis.

Current Daily Workflow

Morning:

claude "Review the N+1 queries in the order list endpoint and suggest fixes"

— scans code, finds lazy-loading in loops, suggests fixes in ~2 minutes (previously 30 minutes via Telescope).

Apply: claude "Implement the fixes and run the test suite" — edits code, runs PHPUnit, reports failures; developer reviews and commits.

New features: Plan → Generate → Review cycle.

Urgent bugs:

claude "The payment callback is returning 500. Check the logs and find the root cause."

— reads storage/logs, correlates with code context, delivers root-cause analysis; developer verifies and decides the fix.

Coding time decreased, but output increased.

Is PHP Still Worth Learning?

PHP demand down 33% is real; junior CRUD roles are vanishing. Yet over 75% of websites still run PHP (WordPress, Laravel, ThinkPHP). These systems need maintenance, upgrades, and AI integration — and developers who can bridge PHP and AI are extremely scarce.

The PHP AI toolchain is no longer empty:

Laravel AI SDK unifies 14 AI providers.

Neuron AI is the first production-grade Agentic framework in PHP.

Symfony AI is iterating rapidly.

Chinese teams are already using ThinkPHP 8 with AI Agents for intelligent customer service, combining RAG with private knowledge bases, and building WeChat mini-program AI assistants with streaming support for all major models. These developers didn't switch languages; they're doing AI work in PHP.

Actionable Starting Point

Stop worrying. Open a terminal and install the SDK: composer require laravel/ai Follow the official docs to create an Agent and connect it to a real need in your current project — even just querying stock or replying to a support message. You don't need Python, microservices, or to abandon your PHP expertise. You only need to shift from "person replaced by AI" to "person who commands AI."

The intern was right: Cursor writes code. But the person who can direct Cursor to write the right code, know what should and shouldn't be written, and debug when things break — that person is still the senior engineer.