Agent traces are the new oil
Summary
Sam Z Liu argues agent traces are the new strategic currency for AI companies, serving as a proprietary intelligence layer that enables model improvement and operational excellence.
Traces differ from conversation transcripts.
While ChatGPT era logs only recorded user-LLM exchanges, agent traces incorporate internal reasoning, tool calls, and agent responses.
The record of user messages, agent internal reasoning, tool calls, and agent responses is an agent trace.
Traces fuel the AI arms race.
Frontier labs use competitor traces to catch up, while data vendors can profit by selling expert-annotated traces to these labs.
The right traces in sufficient quantities make up the backbone of most AI companies today.
Enterprises treat traces as disposable logs.
Most users ignore traces for monitoring or compliance, yet they represent the actual operational work of an agent-driven company.
They are seen as logs, to be filed away for monitoring and compliance
Extracting trace value is difficult.
High token volumes make LLM-based ingestion prohibitively expensive, and unstructured data mixes prose, JSON, and code.
ingesting all of those tokens with yet another LLM is prohibitively expensive
Traces prevent vendor lock-in.
Owning proprietary intelligence derived from traces allows companies to maintain competitive positioning even when using identical base models.
it will be important for every company to own their own intelligence
Highlights
In a world where agents automate most of a business and everyone has access to the same models, your competitive advantage is only as good
The relative simple nature of traces is deceptive. They seem like just a log of what your agent has done, but actually making use of
Read the original ↗Save to your Sponge
Summarized with Sponge. Quotes are the author’s own words.