Tag
6 essays filed under Agent Orchestration.
KV caches aren’t harmless memory. Use provenance-aware eviction and metering to prevent corrupted reasoning, leakage, and runaway spend—with concrete designs.
Read essayApp verification will splinter agent skill distribution. Build signed skill bundles, offline installs, vetting APIs, and fallback runtimes to keep control.
Read essayStop shipping agents on JS/SaaS glue. Compile agents to native, typed runtimes to cut risk, cost, and gatekeeping while improving auditability and control.
Read essayPractical patterns to run multi‑million‑token agents: KV‑cache grafting, partition‑prompt‑aggregate, long‑context replay, and audit‑first memory.
Read essayA practical metering model for agents: bill by work units, not tokens. Prevent leaks, rein in costs, and keep provenance using an enforceable boundary.
Read essayA concrete, testable approach to LLM agent SLAs: specify failure semantics, observability, replanning budgets, and economic duties across tools and vendors.
Read essay