Field notes.
What we learn running millions of messages across eight providers. Mostly about cost, quality and the gap between benchmarks and real traffic.
What multi-model routing actually saves
We modelled two million messages across eight providers at published rates. The savings look real, the mechanism is boring, and the ceiling matters le…
Priya Raman·9 min
The hidden cost of fragmented AI access
The subscription fees are the small part. We looked at what fragmented AI access costs teams in procurement time, unenforced policy and invisible spen…
Jordan Liao·7 min
How we pick a default model, and why it keeps changing
Benchmarks measure what is easy to measure. Here is the evaluation we actually run before changing which model Swurl reaches for by default, and why i…
Marcus Webb·8 min
Subscribe via RSS.