Field Notes archive

Newsletter · Issue 08

Field Notes

The AI news that mattered this week, and what it means for the people doing the work.

Friday, November 20, 2026Covering November 13 to November 195 stories
Get Field Notes every Friday
Jump to this issue's stories ↓

This week

Welcome back to Field Notes. Every week, we pick out the AI stories that matter most to people running real businesses, explain them in plain English, link you to the original sources, and tell you what we make of them.

This week's stories share a question: when an AI agent acts on your behalf, who's responsible for what it does? A business owner lost his Facebook pages for a week after giving his agent access. Banks wrote it into their terms that customers own whatever their agents do. The FBI warned that AI hacking tools are breaking in through forgotten logins in a matter of hours. And Microsoft built an entire product around finding every agent a company runs and giving each one an owner. It's the same question John took on in this week's podcast, and the answer keeps coming back to people.

Here's what happened.

01

A business owner gave his AI agent the keys. Facebook locked him out.

What happened

On November 17, 404 Media reported that Joe Thomas, who owns Kong Koatings Epoxy Countertops in South Carolina, had been using Muse, Meta's own AI agent, to run parts of his marketing, including posting videos to his company's Facebook pages. After he gave the agent access to those pages, Meta's automated security flagged the activity as a likely hacker and disabled the accounts. Thomas relies on Facebook ads for new customers and says the lockout, which lasted about a week, cost him at least $25,000 in work. He got access back after a U.S. senator's office stepped in. On November 16 he filed suit against Meta in Oconee County, seeking $75,000. He says the agent drafted the complaint and he signed it. None of his claims have been tested in court.

Our take

The lawsuit makes the headlines, but the lesson is in what happened before it. A platform's own agent tripped that same platform's security, and a small business lost its main source of leads for a week. Before you give an agent a login to anything your customers depend on, check whether the platform allows agent access, give the agent the narrowest role available (an editor role instead of admin, for example), and make sure a person on your team can still get in if the account gets flagged.

02

Banks are putting customers on the hook for what their AI agents do

What happened

On November 18, TechCrunch reported that Citi had updated its online user agreement. Customers who give a third-party AI agent access to their accounts are now responsible for whatever the agent does, including anything that happens if the tool is hacked or leaks their login. PNC changed its terms the same day, Citizens emailed customers a similar warning, and JPMorgan and Bank of America announced comparable policies last month. Citi says agents it provides itself don't carry the extra liability.

Our take

In Issue 06, we covered banks finding unapproved agents inside customer accounts. Now the rules are in writing, and the risk sits with the customer. The reporting covers personal banking, so if your business banks with one of these institutions, read your business account terms before connecting any AI tool. If anyone on your team has already linked an AI tool to a bank or card account, find out this week and switch it to read-only or disconnect it. For anything that moves money, a person approves every payment.

03

The FBI says AI hacking tools can get from first probe to break-in in under three hours

What happened

In a flash advisory dated November 13 and reported by TechCrunch on November 16, the FBI warned that attackers are using Artex, an open source AI tool originally built for security testing, to run largely automated attacks on law firms, financial services companies, tech firms and others. In some cases, the attacks went from targeting a system to breaking into it in under three hours. The tool mostly finds the basics: weak protections and exposed or stolen passwords. The security company Barracuda disclosed that it was breached after the tool found a login belonging to an outside vendor that Barracuda never shut off when the work ended. In South Korea, attacks on financial firms that lacked multi-factor authentication exposed the personal data of at least 68,000 people.

Our take

Everything this story describes can be defended against with tools you already have. Three habits close most of these gaps. Turn on multi-factor authentication for every account that offers it. Keep a list of every outside vendor with access to your systems. And shut off that access the day a project ends. Attacks now move faster than anyone can watch for them, so these protections have to be in place before something happens.

04

Microsoft's answer to agent sprawl: find every agent and give it an owner

What happened

At its Ignite conference on November 17, Microsoft expanded Agent 365, its tool for tracking and controlling AI agents. Microsoft says organizations have registered more than 50 million agents in it since May. New features find “shadow” agents that employees run on their own computers without IT approval, give each agent its own identity separate from the person using it, and show what AI costs next to how it's used. A feature coming in December will flag agents with no owner and unused agents that still have access. Basic usage reporting for Microsoft's own agents is now included with qualifying Microsoft 365 subscriptions. The full governance features require an Agent 365 license.

Our take

The habit behind Agent 365 works at any size, with or without the software. Keep a simple list of every AI tool and agent your team uses, what each one can reach, and the one person who owns it. Review the list every quarter and shut off anything without an owner. An agent with access and no owner is the kind of opening story 03 describes.

05

84% of business owners use AI. Their biggest skill gap is AI.

What happened

Bank of America's 2026 Business Owner Report, released November 17, surveyed 849 small and 250 mid-sized U.S. business owners in August. Eighty-four percent use AI in some way, most often for marketing (52%), data analysis (41%), workflow automation (33%) and customer service (31%). Among current users, 94% said AI helped their business in at least one way over the past year. Owners expect AI to change jobs more than cut them: among those using or planning to use it, 60% expect employees to move into new or redesigned roles, 52% expect to hire people to manage AI tools, and 33% expect it to reduce headcount. AI and emerging technology was the most-cited skill gap among employees, at 36%. Meanwhile, costs went up for 91% of owners.

Our take

The skill-gap number is the one to act on. Half of owners named AI tools as their top digital investment for next year, and training rarely makes it into that budget. If AI is on your list for 2027, set aside time and money for the people who'll use it, and pick one workflow where you can measure the before and after.

This week from Avolis

If something in this issue sparked a question about your own business, we'd love to hear it. Reach us at avolis.ai/contact.

See you next Friday.

The Avolis Team

Field Notes, every Friday

The AI news that mattered, in your inbox.

A short weekly read on the AI stories that affect real businesses, explained in plain English, with our take on each.

Ready to make AI work for you?

Book an AI readiness evaluation. If there’s nothing worth automating, we’ll tell you.