Mihai Kondort

Software developer with over 20 years of experience. I have worked across desktop applications, data engineering, full-stack web development, and native mobile apps on both iOS and Android. I am not tied to any one technology or language. They are tools for solving problems and delivering results.

I spent over a decade in the financial sector, working in Data, ETL, and Business Intelligence at ING, Ingenico ePayments, AIB Bank, and IBM. Most recently I delivered full-stack, desktop, and mobile work at Eightpoint, where I led a development team and built an AI workflow orchestrator that partly automated a code migration from iOS to Android. I have lived and worked in four countries: Romania, England, Ireland, and the Netherlands. That has made me quick to adapt to new teams and new ways of working. I have experience in team leadership, mentoring, and delivering across the whole product lifecycle.

I use AI in a planned way, not ad hoc. Every feature or fix starts as a written analysis document. There I refine the idea and, where it matters, research current industry practice before I commit to an approach. Once the analysis holds up, it becomes an implementation plan. All the technical work is broken into phases, and each phase into concrete tasks.

Each phase opens with a P0 re-validation task. It checks whether earlier phases, and any changes made along the way, have invalidated what comes next. The plan is corrected before any more of it is executed. Progress is tracked inside the plan itself and updated as each task completes, so interrupted work can resume in a new session with nothing lost.

I mix the tooling on purpose. Claude Code drafts the analysis, then the plan, then the code, but nothing moves to the next stage until I have worked through it myself. I challenge the analysis and correct it, and I only approve a plan once it matches what I actually want built. Codex then reviews the same work as an independent second opinion, the analysis and the plans as well as the code. A second model criticising the first catches far more than either does alone. The models do the work that can be automated. The judgment about what is worth building, and whether the result is right, stays with me.