Auto doesn't mean smart
I’ve tried auto-routing in every major harness I use regularly. Cursor, Copilot, Windsurf, Claude Code. The result is the same every time: it picks wrong. Not occasionally — con... Read →
Engineer at the bleeding edge of AI, battle-tested at enterprise scale.
I'm David Morgan. I build production AI systems in AT&T's Chief Data Office, plus a steady stream of side projects on my own time. This is where I write about working with agents, and link to the things I've built.
The flywheel is spinning up: quantum, AI, and the convergence nobody priced in
Five years ago, if you’d told a room full of serious people that AI would be writing production code, winning Nobel-adjacent science prizes, and forcing companies to rewrite the... Read →
We have no way to measure AI productivity
We are building AI systems faster than we can measure what they’re worth, and burning compute faster than we can produce it. Those two gaps are about to collide, and when they d... Read →
Strategical Context Injection: the unifying frame for skills, workflows, prompts, and .md files
I switched between three AI coding tools last year, and somewhere in the middle of it I stopped having to relearn anything. Read →
The bottleneck has moved
Colleagues come to me a lot. The conversation almost always goes the same way. Read →