Weekly Signal #42
AI
LLM-as-judge quietly rewards length — three calibration fixes that actually move agreement.
Growth
The infra bottleneck moved from compute to interconnect; capex guidance is a lagging tell.
Humanity
What stays human when the agent does the drafting: taste, judgment, and knowing which signal to follow.
This week leaned toward reliability. The judges we trust to grade AI turn out to need grading themselves, and the constraint everyone underwrites — compute — has already handed the baton to power and interconnect.