An AI tool drafts a reply to a customer. Fine. Now let it open the account, then issue the refund. Same chat window, very different job.
In this episode I walk one customer message through three steps, the draft, the account and the refund, and show where AI goes from helpful to dangerous. I also look at what “a human reviews it” should actually mean, why a wrong AI answer can cost you more than the same mistake from an employee, and what I’d do tomorrow instead of banning AI.
Chapters:
0:00 Riding along
1:23 Draft
1:47 Account
2:04 Refund
2:16 Where I draw the line
3:42 What “a human reviews it” means
4:59 Why mistakes cost more
5:58 Embrace it
7:16 Vet the vendor
8:39 What to do tomorrow
Sources:
UK NCSC, agentic AI blog, 20 Aug 2026
OWASP LLM06:2025 Excessive Agency
Mata v. Avianca, S.D.N.Y., 22 Jun 2023
Dietvorst, Simmons and Massey, 2015, algorithm aversion
Microsoft and LinkedIn, 2024 Work Trend Index
IBM Cost of a Data Breach Report 2025
Anagram survey via HR Dive, 13 Aug 2025
DHS order on DeepSeek, Federal Register 91 FR 66346, 20 Oct 2026
I’m David Shaw, a vCISO and the founder of Corvus Cybersecurity