Laravel Workflow is a powerful tool for orchestrating long-running, stateful workflows in PHP. Paired with PrismPHP, it becomes a compelling foundation for building reliable AI agents that not only generate structured data but verify and retry until results meet strict real-world constraints.
In this post, we’ll show how to use Laravel Workflow + Prism to create an agentic loop that:
- Generates structured data using an LLM
- Validates the result against custom rules
- Retries automatically until the result passes
You can try this exact workflow right now in your browser with no setup or coding required. Just click the button in the Laravel Workflow Sample App and launch a GitHub Codespace to run it.
What We’re Building
We’ll create a workflow that asks an LLM to generate a user profile with hobbies. Then we’ll validate that:
- The name is present
- At least one hobby is defined
- The name starts with a vowel
If the result fails validation, we loop back to the LLM and regenerate. All of this is durable, asynchronous, and tracked through stateful events.
Step-by-Step Example
- Console Command to Trigger the Workflow
use App\Workflows\Prism\PrismWorkflow;
use Illuminate\Console\Command;
use Workflow\WorkflowStub;
class Prism extends Command
{
protected $signature = 'app:prism';
protected $description = 'Runs a Prism AI workflow';
public function handle()
{
$workflow = WorkflowStub::make(PrismWorkflow::class);
$workflow->start();
while ($workflow->running());
$user = $workflow->output();
$this->info('Generated User:');
$this->info(json_encode($user, JSON_PRETTY_PRINT));
}
}
- Define the Workflow Logic
use Workflow\ActivityStub;
use Workflow\Workflow;
class PrismWorkflow extends Workflow
{
public function execute()
{
do {
$user = yield ActivityStub::make(GenerateUserActivity::class);
$valid = yield ActivityStub::make(ValidateUserActivity::class, $user);
} while (!$valid);
return $user;
}
}
This is a classic agent loop. If validation fails, we prompt again automatically.
- Generate Structured User Data with PrismPHP
use Prism\Prism\Prism;
use Prism\Prism\Enums\Provider;
use Prism\Prism\Schema\ArraySchema;
use Prism\Prism\Schema\ObjectSchema;
use Prism\Prism\Schema\StringSchema;
use Workflow\Activity;
class GenerateUserActivity extends Activity
{
public function execute()
{
$schema = new ObjectSchema(
name: 'user',
description: 'A user profile with their hobbies',
properties: [
new StringSchema('name', 'The user\'s full name'),
new ArraySchema(
name: 'hobbies',
description: 'The user\'s list of hobbies',
items: new ObjectSchema(
name: 'hobby',
description: 'A detailed hobby entry',
properties: [
new StringSchema('name', 'The name of the hobby'),
new StringSchema('description', 'A brief description of the hobby'),
],
requiredFields: ['name', 'description']
)
),
],
requiredFields: ['name', 'hobbies']
);
$response = Prism::structured()
->using(Provider::OpenAI, 'gpt-4o')
->withSchema($schema)
->withPrompt('Use names from many languages and vary first initials.')
->asStructured();
return $response->structured;
}
}
- Validate Business Logic
use Workflow\Activity;
class ValidateUserActivity extends Activity
{
public function execute($user)
{
if (empty($user['name']) || !is_array($user['hobbies']) || count($user['hobbies']) === 0) {
return false;
}
foreach ($user['hobbies'] as $hobby) {
if (empty($hobby['name']) || empty($hobby['description'])) {
return false;
}
}
// Extra Validation: The user's name must start with a vowel.
if (!in_array(strtoupper($user['name'][0]), ['A', 'E', 'I', 'O', 'U'])) {
return false;
}
return true;
}
}
What Makes This Pattern Powerful
This design pattern is what you’d call a reliable agentic loop:
- LLM generation via Prism
- Validation & retry via Laravel Workflow
- State persistence for crash recovery or inspection
- Observability via Waterline
It’s perfect for AI applications where accuracy, safety, and traceability are required.
Try It Now in Your Browser
We’ve bundled this workflow into the official Laravel Workflow Sample App, which runs in GitHub Codespaces.
To launch it:
- Open the sample-app repo
- Click the “Code” button → “Codespaces” → “Create codespace on main”
- Wait a few seconds for setup
- Set your OPENAI_API_KEY
- Run:
php artisan migrate
php artisan queue:work
- In a second terminal:
php artisan app:prism
You will see the queue working and eventually see the validated output.
Where to Go From Here
You can easily adapt this pattern to:
- AI agents for form filling
- Data scraping and validation
- Content generation with retry policies
- Moderation and review queues
Each step remains reliable and traceable thanks to Laravel Workflow’s durable execution model.