September 28, 2026
What people mean by “agent”
The word covers a tool call, a loop, and a person reviewing a draft. Those are not the same product.
Read →
Notes
A few notes on the words, the measurements, and the questions that show up when people put models into real work.
September 28, 2026
The word covers a tool call, a loop, and a person reviewing a draft. Those are not the same product.
Read →
September 18, 2026
A larger window is not automatically a better request. Extra text costs money and can bury the instruction.
Read →
September 7, 2026
Swap models only if you can name a real task that got better. A benchmark moving is not that list.
Read →
August 26, 2026
Latency, a day of real traffic, and how often a person has to fix the output. Demos skip all three.
Read →