Field Note · Part 3 · June 5, 2026
A capability resolution layer should return more than a tool name. It
should return a structured decision package that agents, platforms, and
operators can trust.
Field Note · Part 2 · May 1, 2026
Wrong tool use is not just a correctness problem. In agent systems, it can
create retry loops, wasted model calls, higher latency, and weaker AI
economics.
Field Note · Part 1 · April 10, 2026
As agent systems gain access to more tools, APIs, workflows, and
sub-agents, the harder platform problem moves upstream: resolving the
right capability before execution.