Why Resolve AI: Token Efficiency at Scale

Token usage has become a scoreboard across the industry, largely because nobody agreed on how to evaluate what AI actually delivers. Varun makes the case that this is the wrong measure entirely. Tokens only represent how hard a model is working. What matters is the outcome on the other side.

In this video he defines token efficiency as the number of tokens burned to reach an outcome, whether that is investigating an alert or getting to a root cause, and explains why Resolve AI focuses on it: to keep cost predictable and quality high at the same time.

The place that game is won or lost is context. Varun uses a detective analogy to explain it. A dumb detective takes the first answer and closes the case. A nervous detective interviews the whole town and never reaches a conclusion. A smart detective asks the right questions and works across code, infrastructure, and telemetry to reach a precise answer.

Plotted out, agent quality against context looks like a hill, with low quality at both extremes and a peak in the middle. Getting to the top of that hill is a research problem, and it is the reason Resolve AI charges customers on outcomes rather than tokens.

Resolve AI is the first agentic interface for engineers to operate their production systems. Using natural language, engineers can work across their code, infrastructure, telemetry, and team knowledge all in one place.