Alex Rezvov
Systems Developer & Technical Writer | Specialist in Resilient Agent Operations
Sources checkedAlex Rezvov is a software engineer and technical author at blog.rezvov.com. Alex has spent years investigating real-world edge cases in autonomous software systems, writing widely cited engineering post-mortems on multi-agent deadlock, token context exhaust, and automated supervisor watchdog daemons.
Areas of focus
Professional niches
Proof of Work
Post-Mortem: 16 Production Agent Failures & The Watchdog Daemon Architecture
Empirical study analyzing 16 distinct failure modes in unattended production agents (including context overflow loops, API timeout deadlocks, and stale session states) with an open-source supervisor watchdog.
Supervisors that blindly restart crashed agents without backoff can create API throttling cascades if upstream services are degraded.
Context: Node.js, Python, systemd watchdog, SQLite state logs, and LLM telemetry.
View mission