Debate and philosophy taught me to make an argument clear. Building systems made that clarity practical.
Although I’m not contemplating the big questions of epistemology, metaphysics, philosophy of mind, and ethics directly as often as I did as a student, these areas have remained very salient in my work and life.
Years of competitive debate and then a Philosophy, Politics, and Law degree taught me how to take a position apart, find the premise everything rests on, and rebuild the case so that someone who disagrees can still follow every step.
At INFAMOUS, inventing the services meant living their costs. A large percentage of every week went to compiling evidence of finished work, personalizing hundreds of pitches by hand, and manually deduplicating in/out-of network curators while researching pitch targets. The first instruments I developed were small automations: keyboard and browser macros, text snippets, and mail merge, but it was the first time I understood the shape and feel of the work was itself something I could iterate and improve on.
A defining moment was handing off the (at this point largely still manual) process to a new employee — a remote assistant with no music background and English as his second language. This exercise forced me to define the process better than I ever had for myself. Every step, edge case, and condition that felt obvious to me had to become a clear instruction for someone else to follow reliably.
Founding and running my own businesses has compounded this. The low-context executor became an AI agent as often as a person, and agent-driven automation matured into a real operating advantage, built to my own judgment and taste.
That advantage became a second practice. Consulting is doing for other teams what I did for my own operation: turning fuzzy workflows into explicit states and owners inside the tools people already use.
This training combined with my educational background turned out to be the right preparation for working with AI. The work has to be specified from first principles, with the goal stated, the constraints named, the edge cases enumerated, and a definition of done that someone with no context can check. I use deterministic rules and tests where possible, because an executor who has to guess produces work that has to be audited.
Debate and Philosophy also trained me to look for the second-order consequence, the thing hidden in the text that is often unnoticed, which could produce unwanted results. Writing instructions for agents, and editing the outputs produced by agents, is exactly that. The checklists I once wrote for a human assistant now go to agents.