Without evals, every change is a guess
Why an AI agent needs evals before its next prompt or model change, what makes an agent harder to evaluate than a single LLM call, and the task, run and grade loop every eval is built on.
Hi, I’m Roberto. I build things with Laravel, and with solder and wood.
Notes on PHP packages, home automation, synthesizers and DIY, written as I go.
Latest posts
See all →Who judges the input: screening prompt injection with Laravel Judgment
Putting a separate judge in front of an LLM-based CV screener: Laravel Judgment asks Jev whether a CV contains instructions aimed at the machine, and the application decides what to do with the answer.
When you can't write the rule: Laravel Judgment with Jev and Laya
How to let a model assess free-text refund requests with Laravel Judgment while the application keeps the decision, first on the hosted Jev and then on a self-hosted Laya server.
Content is not an instruction: prompt injection countermeasures in Laravel AI
How a line of hidden text in a CV can steer an LLM-based screener, and the countermeasures from LLM Application Engineering that limit it: separate instructions, checks before and after the call, least privilege and approval gates.
Create a custom validation rule with string alias
In this tutorial I will show how to define a Laravel validation rule providing a custom alias string to use in custom validation classes.
Books
See all →
Laravel 13: The Unwritten API
An open-source book on the public, usable but undocumented parts of Laravel.
Laravel Testing: Deploy without fear and without compromise
Backend testing in Laravel, from feature tests to CI and mutation testing, with Laravel 13 and Pest 4.
LLM Application Engineering: Principles for reliable systems, illustrated with Laravel AI
The decisions an LLM-based application has to get right before it holds up in production, illustrated with Laravel AI.
Latest video
See all →
Glowing LED Christmas tree star
The Christmas LED topper glowing on top of the tree. Music: "Stacks Of Stuff" by Simon Panrucker and "Little Star" by Manuele Atzeni.
Next talk
See all →International PHP Conference Munich 2026 · Munich, Germany
Less Telepathic Codebases: AI-Readability for Laravel Applications
AI agents are now part of many developers’ daily workflows: they read repositories, answer questions about code, suggest changes, and support review activities. But having access to a project’s files does not mean truly understanding its behavior. Many Laravel applications are “telepathic”: they require implicit or unwritten knowledge to be understood. Domain rules are scattered across controllers, models, policies, jobs, listeners, and tests; architectural decisions are never documented; names are generic; conventions are known only to the team. This talk introduces AI-readability as an architectural quality: making a codebase more explicit, inspectable, and verifiable by AI tools as well. We will see how to unearth the knowledge buried in Laravel artifacts — routes, models, policies, jobs, events, listeners, commands, and tests — and turn it into structured context: application manifests, readability audits, source-grounded prompts, architectural diffs, and ADRs inferred from code and tests. The problem is not that AI cannot read the team’s mind. The problem is that too many codebases require it to.