Чем предстоит заниматься
System & Platform Design: Own infra topology, scaling and blast-radius boundaries, the build/deploy graph, database ownership boundaries, and cost ceilings — the decisions agents execute but can't make
Guardrail Engineering: Build the validation gates, idempotency checks, drift detectors, and runbooks that keep both humans and agents from doing damage. Your deliverable is the safety system, not the deploy
Escalation Tier: Diagnose the novel failures — consumer lag, DLQs, materialized-view chains, consent ordering. Write the runbook the first time; agents handle it after
Reliability, Capacity & Cost: Own the 500K events/sec target: load testing, Karpenter/HPA tuning, Database sizing, SLOs and alerting
Security & Secrets Boundary: Own IAM/IRSA, secret rotation, the auth chain, and what the agent/MCP surface is allowed to do
Agent-Fleet Operability: Keep skills, runbooks, and agent context in sync with reality. A stale runbook is a wrong actor, executed confidently
Incident Leadership: Lead incidents and capture every fix as a new runbook plus regression test