Six AI coding agents. One production task. The results reveal a gap that no leaderboard can see. I've spent years evaluating technology for a living — first in physics labs measuring things at 10⁻¹¹ torr, then in venture doing technical due diligence on blockchain protocols, now advising on AI
There is a marvelous paradox at the heart of enterprise software evaluation, which is that everyone involved knows the process is performative nonsense, yet everyone continues to execute the performance with utmost sincerity, like method actors who've forgotten they're in a play. Last week I conducted
I have a custom app that plans my workouts based on my family calendar, my work schedule, and my energy patterns. I didn't buy it. I didn't hire a developer. I built it myself, in an afternoon, with AI. And I'm not a developer.
Steve Jobs didn't trust touchscreens. Well, that's not quite right. Steve Jobs championed touchscreens while most of Apple internally thought he was insane. User research said touchscreens felt wrong. Experience said they wouldn't work. And Jobs said: do it anyway. Here's what
What Would You Build If Labor Cost Almost Nothing? I was in a room with corporate executives last week, talking about AI and the future of work. Everyone had the usual talking points ready: automation, efficiency, unlocking unstructured data, faster decision-making. I realized we were all missing the plot. We&
2025 was the year I stopped writing code and started orchestrating it. The productivity delta wasn't marginal—it was categorical. By treating AI as a capable operator rather than a sophisticated autocomplete, I shipped: a family-specific child development tracker, an AI-driven workout system that accounts for schedule conflicts
Documentation decays. Text files describe how a system should work, but code describes how it does work. When the two diverge, code wins. This is not a moral judgment; it is a fact of execution. The system runs the code, not the documentation. But what if documentation were executable? What
Security is often presented as a checklist: encrypt data at rest, use HTTPS, implement authentication, configure firewalls. This framing suggests security is a state you achieve by completing tasks. But security in deployed systems is not a state; it is a process. Each layer builds on the previous, and the
Standard practice in web application deployment is a three-tier environment: development, staging, and production. Development runs locally or on a dedicated server. Staging mirrors production configuration for pre-deployment testing. Production serves live traffic. This pattern provides isolation and a safety net — if staging deployment fails, production remains untouched. But standard
There is a tension in software systems between designing architecture and deploying infrastructure. Most teams design first, then deploy. This sequence feels safe: you know what you need before you provision it. But this safety is illusory. Infrastructure decisions encode architectural assumptions, and those assumptions remain invisible until you attempt
At this point in time, you are in a certain state. Your digital reading list spans dozens of newsletters, your Kindle library could fill a warehouse, and your browser tabs have tabs of their own. Maybe you are satisfied with your information consumption. Maybe you are not. But like all
At this point in time, you are in a certain state. Your lean team is deploying AI with varying degrees of success. Maybe you're happy with the results. Maybe you are not. But like all things in life, the only certainty is the uncertainty of your evolution toward