Research
State of inference
Our plans for research that helps businesses choose AI models, decide where to run them and understand when to switch.
Reference data as an engineering advantage
How purpose-built reference tools give an agent the information it needs, and how they compare with general SQL access on reconciliation tasks.
The cheapest model is not the cheapest model
Testing model configurations on a production agent’s real tasks to understand the tradeoffs between inference cost, accuracy and completion time.
When does owning inference pay?
When workload volume, hardware utilization and the need for control make owning inference infrastructure worth considering.
The token is not the unit of intelligence
Why token prices and benchmark scores do not tell you what useful work will cost.
Crown chasers and price breakers
How competition between frontier models and cheaper challengers changes the choices available to the businesses using them.