This week's four stories all come back to trust: a familiar voice on the phone, an agent that can spend money, an assistant that agrees with everything you say, and voters who are uneasy about how fast AI is moving. Here's what happened, why I think it matters for your business, and what I'd do about each one. If you want a second set of eyes on any of these, talk with the Spej team.
AI impersonators allegedly stole more than $129 million from companies.
On November 5, the U.S. Attorney's Office for the Southern District of New York unsealed charges alleging that three people used AI to mimic executives and stole more than $129 million from companies between July 2024 and February 2026, including by directing wires through someone posing as outside counsel. These are charges, and the defendants have not been convicted. Require a callback to a known number for any wire or bank-detail change, and tell accounting you will never be upset that they checked.
AI agents show back-office results while shopping on customers' behalf remains early.
Visa's November 5 benchmark of 22 models failed 42.8% of shopping test runs, and about 54% of those failed runs still ended in a purchase. Uber reported about 500 Eats orders and four rides from third-party agents in a week of about 320 million trips, while Airbnb said about 40% of support issues in the quarter were resolved without a human and support cost per booking fell more than 17%. Start with a bounded task you control, record the current effort, and put a spending limit and a confirmation step on any agent that can pay.
Agreeable AI can weaken the decisions leaders need it to improve.
A November 4 Harvard Business Review article described a BT Group trial of 2,400 people in which an AI tool built to push for more specific goals produced three and a half times as many excellent goals as default AI or no AI. The limit matters: one company, one task, reported by the people who ran it. For pay, hiring, performance, layoffs, and pricing, have the responsible person write their own view before they see the AI recommendation.
Voters backed limits on data center development across local ballot measures.
Ballotpedia reported on November 6 that voters favored limiting data center development in 40 of 41 local measures on November 3, across 10 states, and an NBC News exit poll found 76% would oppose a new data center in their community. Ask your main AI vendors where your data is processed and how they would handle new state requirements, and be specific internally about which work AI handles, which work it does not touch, and who remains responsible.
First published on Spej's Substack.
