SoftBank tests AI agents that run parts of its phone network on their own

How to do it yourself

Our suggested steps. The company didn't publish theirs.

1. 1

   Set up uptime and error alerts for your site, POS or key systems.

2. 2

   Send alerts into n8n.

3. 3

   n8n asks Claude (prompt below) to diagnose using recent logs and your runbook.

4. 4

   Claude posts a diagnosis and suggested fix to Slack; a human approves any change.

5. 5

   Add the fixes that worked to the runbook.

The instructions to give the AI

Paste this into Claude (the free plan is fine), swap the [brackets] for your details, and test it on a few real examples.

This is one part of the setup. It won't connect to your phone, inbox or other apps on its own; the steps above cover that.

You are an on-call engineer for [business]. Alert: [paste]. Recent logs: [paste]. Runbook: [paste]. 1) Most likely cause, 2) evidence, 3) safe first step to fix, 4) whether it needs a human now (yes/no) and why, 5) message to post to staff.

What you should get back: "Likely cause: disk full on db-02 (logs show 98%). Safe first step: clear old logs per runbook 4.1."

Written by Quickly and Good, inspired by this case.