Analytical reports, operational frameworks, and research notes from WBA's independent practice.
Thirty days of recorded tool calls from a live instrumented system: 121 tools, 6,406 calls, an 83.3% tool miss rate. What the record shows about measuring epistemic debt rather than arguing about it.
Analysis of how persistent context — files, documentation, and structured workspaces — functions as the critical infrastructure layer separating Stage 1 AI usage from genuine operational integration.
The practical value of a secondary model is not hype. It is preserving premium context by offloading broad scans, summaries, and first-pass analysis.
Serious AI use stops asking which model is best and starts asking which model is best for this step of the workflow.
When AI usage feels suddenly more expensive, the assumption is that models have grown more capable and therefore more costly. The data tells a different story. The cost is not reasoning. The cost is the environment failing to support the agent the model has already become.
