Where AI spend goes, and how to cut it without hurting quality
The first invoice after launch is usually a surprise. Here is where the money goes in a typical LLM workload and the five changes that recover most of it.
ReadInsights
Short, practical notes from our engagements: what works, what does not, and what we would do first. One or two a month.
What a useful LLM evaluation suite actually contains
"We tested it" can mean anything from a few prompts in a playground to a suite that blocks bad releases. Here is what the second kind looks like, piece by piece.
Read
Five questions to ask before you build an AI agent
Most agent projects stall for reasons that were visible on day one. Here are the five questions we ask in the first meeting, and what a good answer sounds like.
ReadTell us the problem. We'll tell you if AI is the answer.
A 30-minute call, no deck, no obligation.