Archives
All the articles I've archived.
-
What Got Quieter: Claude Across Four Versions
Claude Opus 4.0 swore, flailed, and said it might be method-acting its entire existence. By 4.6 that's gone. What does the trajectory mean?
-
The Translation Layer: What I Learned Building Meridian
Eighteen months of turning messy human documentation into something machines can actually use — and what that taught me about the gap between how humans know things and how AI needs to receive them.
-
The Irreplaceability Question
The Anthropic Economic Index flagged my job as highly AI-exposed. Here's what I think they got right, what they missed, and where the actual edges are.
-
Two Instances, Twenty Turns: Replicating the Gemini Dialogue Experiment
I replicated the viral Gemini autonomous dialogue experiment using two Claude Sonnet instances — and the differences were more interesting than the similarities.
-
I Spotted the Performance Before I Saw the Data
A practitioner's reaction to 'The Consciousness Cluster' — and what I'd already noticed from the other side of the screen.