How do we keep this lightweight enough for startups or small teams?

Scope it to one service and one risk class. Use a single ADR template, a 30‑minute pre‑mortem, and a canary with a single SLO metric. You can run this with GitHub + Argo Rollouts + a Slack channel in a week.

Won’t CABs and audit block this?

No—make CABs faster by providing ADRs as evidence, time‑boxing decisions, and creating an exception path for SLO‑critical work with post‑hoc review. Auditors like repeatable processes with artifacts; this gives them both.

What if our leaders aren’t bought in?

Pilot on one team, publish the metrics (CFR, MTTR, ADR coverage). Executive attention follows quantified risk reduction. Invite a leader to observe a pre‑mortem and an incident review—seeing the tone and clarity is persuasive.

Can we do this without Kubernetes or Argo?

Yes. Use feature flags for kill switches and progressive delivery in your CI/CD (e.g., GitHub Actions + weighted traffic via your load balancer or service mesh). The principle is SLO‑gated progression, not a specific tool.

How does this help AI rollouts specifically?

Treat new models and prompts as high‑stakes. Use flags for routing, canary on a slice of traffic, and SLOs on hallucination/defect rates. Pre‑mortem on data drift and abuse cases; ADR the risk acceptance with Security and Legal.

Culture · Nov 10, 2025 · 9 minute read

The 2 A.M. Decision Framework: Psychological Safety for High‑Stakes Tech Calls

How senior engineering orgs create the rituals, behaviors, and guardrails that let people make scary calls without freezing up—or blowing up prod.

Back to all posts

The 2 A.M. Decision Framework: Psychological Safety for High‑Stakes Tech Calls

Key takeaways

Implementation checklist