Case study

Day-2 operations in natural language across a growing fleet

Atlas Analytics (placeholder) · B2B analytics

Challenge

As the fleet grew past a handful of hosts, junior operators hesitated to run production commands without senior pair-programming.

Intent in English, verification on the host—not folklore in a channel.

Architecture

  • 8 application VMs
  • Per-server TESS chat contexts
  • Mixed SSH and agent connectivity

Solution

Natural language ops with plans, live streams, and verification—scoped strictly per server.

Implementation

  1. 01Standardized host onboarding with auto status check
  2. 02Playbook phrases encoded as recurring intents
  3. 03Read-only for contractors; always-approve for core platform team

Before / after

Before

Slack runbook + senior-on-call

After

NL plan + approval + evidence

Deployment workflow timeline

  1. Quarter 1

    Pilot NL ops on two services

  2. Quarter 2

    Fleet-wide onboarding checklist

Results & metrics

  • · Juniors complete restarts and config checks independently
  • · Seniors review audit-style history instead of pair-ssh
  • −6h

    Senior pair hours / week

  • 71%

    Ops tickets closed in chat

Lessons learned

  • Context isolation beats a single global chat for production safety.

Future plans

  • Org policy packs for contractors

Discuss a similar rollout

We help teams design approval, verification, and multi-host operations with Kairo.