Architecture decisions, incident post-mortems and policy patterns from the team building the firewall for autonomous agents.
Every millisecond your policy engine adds is a millisecond your agent stalls. How we keep warm-path decision latency under 15ms while still evaluating cost, risk and compliance in parallel.
Engineering · 8 min read · Sep 9, 2026
Instead of asking “what can this agent do?”, ask “what happens when it does the wrong thing?” A practical framework for scoping tool access.
Security · 6 min read · Sep 2, 2026
Plain-language rules are only useful if they are unambiguous to a machine. How we compile natural-language policies into deterministic decision graphs.
Product · 5 min read · Aug 26, 2026
Treating injected instructions as a permissions failure — not a prompt-engineered content filter — changes the whole architecture.
Security · 7 min read · Aug 19, 2026
Millisecond-level, tamper-evident records of every tool call: what to capture, what to redact, and how long to retain it.
Engineering · 9 min read · Aug 11, 2026
Shadow mode, dry-run, human-in-the-loop, then autonomous. A four-stage path that lets you ship capability without shipping risk.
Product · 6 min read · Aug 4, 2026
One email a month: architecture decisions, incident post-mortems and policy patterns. No product spam.
