How I lead engineering

Organizations

• Build the team before the roadmap

  Hiring, leveling, and mentoring decide whether strategy survives contact with reality. I hire for judgment and ownership, then give people room to lead.

• Mentorship is a delivery system

  Staff and senior growth isn’t optional soft work—it’s how an organization compounds. Clear expectations, real ownership, and honest feedback beat performative mentoring.

Technical strategy

• Architecture reviews that decide

  Reviews should force a choice: options, risks, and an owner—not slide decks that end in “let’s discuss further.”

• Architecture should evolve

  Prefer evolutionary architecture. Temporary dual stacks are tools; permanent accidental complexity is a leadership failure.

• Bias toward simplicity

  Add complexity only when the problem demands it—and remove it when it no longer does.

Delivery excellence

• Build incrementally

  Avoid big-bang rewrites. Ship surfaces, learn from production, and retire legacy on evidence—not optimism.

• Measure before optimizing

  Use delivery signals—DORA-style where it fits—before declaring a process or architecture “better.”

• Product over technology

  Technology is a means. Business outcomes and customer value decide whether an architecture earned its complexity.

Decision making

• Documents over meetings

  Write decisions down. Clarity compounds; status meetings rarely do.

• Own the outcome

  Own the outcome, not just the implementation.

  Solve the problem, not just the ticket.

• Close the loop

  Follow through. Communication isn’t complete until Product, Design, QA, and Engineering know the outcome.

AI adoption

• AI augments engineers

  Automation should improve developer experience and speed—not replace engineering judgment.

• Platform before demos

  Voice and Agentic AI only matter when reliability, evaluation, and product integration are treated as platform work—not one-off experiments.

• Customer problems first

  Whether the stack is React, Voice AI, or agents—value only counts when it solves a real customer problem.