Vol.045 — Your Old System Was Built for Human Mistakes. Your AI Agent Doesn’t Make Them.
Date: 2026-06-16 | Newsletter
Key Summary
In Vol.045, Zenta reads Salesforce’s $3.6 billion acquisition of Fin — the AI customer-service agent formerly known as Intercom — in the same week it trimmed its in-house Agentforce team, and refuses the easy “even the vendor doesn’t trust its own AI” take. His angle: the SaaS your team logs into every morning — MYOB, Xero, the Salesforce form — was engineered to catch one thing, human error. Mandatory fields, dropdowns, approval gates, double-entry: every one is a wall around a human weakness. An AI agent never forgets step four and never has a bad morning, so it sails through those gates — while making an entirely different mistake nobody is watching for: misreading a number, a date, the wrong column, and acting with total confidence on something it doesn’t actually understand. The tip for owners adopting AI this year: match the check to the worker. Verify what the agent read, and fence it to what it genuinely knows. For ASEAN SMEs the trap is specific — drop an agent into familiar software, trust it because it looks familiar, and never build the new checks. Do it simple. Stop supervising your AI worker like a human – it fails differently, so you have to check it differently.