Page 4 of 6
Thoughts on AI engineering, responsible automation, and building systems that matter.
Retries are easy until an agent sends the same email twice, creates two tickets, or charges for one request more than once. Reliable automation begins by making repeated execution safe.
Read more →
Laravel gets called magical when it hides work developers expected to see. After years of building PHP systems, I think the real question is not whether Laravel has magic. It is whether your team knows where the magic ends.
Read more →
An AI feature can look impressive in a request-response demo and still collapse in production. Queues turn slow, unpredictable model calls into workflows that can be retried, inspected, limited, and trusted.
Read more →
I needed to build a CLI tool that could orchestrate Docker containers, manage Git worktrees, allocate ports, and run a 9-stage pipeline in under a minute. I evaluated five options. The winner surprised me.
Read more →
The first time a lawyer asked me "but who is liable when the AI makes a wrong decision?" I realized I had been building systems without understanding the world they operate in.
Read more →
The demo is usually the easiest part of an AI agent project. The harder work starts when the agent has users, tool access, schedules, memory, logs, failures, and a growing list of edge cases that nobody mentioned during the prototype.
Read more →
People often ask what my actual setup looks like now that my work sits between PHP, enterprise systems, AI agents, and solo automation. The short answer is that I optimize for reliability, speed, and context retention. Not minimalism for screenshots, not endless productivity gadget collecting. Just a stack that lets me move fast without losing control.
Read more →
A dashboard can tell me that an agent failed. A useful event trail tells me which decision, tool call, approval, or external dependency caused the failure — and whether the system recovered safely.
Read more →