Process
How I work
Five steps, in this order. The one that decides the outcome is the first, and it is the one everybody skips.
Understand the problem
What is failing, who it hurts and what it costs. On the real-estate aggregator the brief said “cut costs”, but the actual problem was depending on someone else for the core of the product. Those are two different problems and they lead to different answers.
Design the solution
Architecture, data model, dependencies and what each option gives up. This is where I decide what a failure is allowed to take down with it, what can safely be retried, and what has to be transactional.
Build it
Code, infrastructure and deployment. When a process can fail, I try to make it fail in isolation and be safe to retry: one broken portal should not stop the other six.
Validate
Real behaviour and edge cases, not just a green build. A backup does not count until you have restored one; a deployment does not count until the healthcheck answers.
Measure and adjust
Cost, timings and maintenance. If I say something got cheaper or faster, I have the number before and the number after — not an impression.
Where AI fits
In the steps where it adds speed, never in the ones that decide the design. The technical decisions and the validation stay mine, and I answer for them.
How I use AI, in detail