John Matoso is back for the second Coffee Brief. He owns SubTonique. This one was live again, and we covered more than I planned.
We opened with something that happened that week. Anthropic announced a new policy about abusive behavior toward Claude, and I wanted John’s take on it. That turned into a conversation about how we talk to AI, and then into the thing I care about more: owners are buying AI tools and not teaching anyone how to use them.
Does how you talk to AI show up in how you lead?
A correction first. On the episode I said Anthropic was going to ban users who are abusive to Claude. I did not have the criteria, and I said so. I looked it up afterward. The policy was announced on October 8 and takes effect November 12. It covers sustained and needless abusive or cruel behavior, and Anthropic says it is meant for extreme cases. Claude ending the chat is the main enforcement they describe. I could not confirm account bans, so read my “ban” on the episode as me being loose.
John came at it from the operations side. He was struck that the language had to be spelled out at all. He also thought about the cost. If people at scale are hurling abuse at a model, that is a lot of tokens spent on an argument that goes nowhere. He thinks anthropomorphizing AI is risky in itself, and he wondered whether abusive chats on free models could end up in training data. He was clear that we have no indication of that. It was his guess.
My angle was different. A lot of AI is designed to keep you coming back, the same way social media was. If it always tells you that you are great, and your real friends and coworkers sometimes complain, which one do you want to chat with? Then there is the habit. The way you speak to someone in private tends to be the way you speak in public. If you are a kind person in private, you are probably kind in public. So I do not think it is really about the AI. It is about whether you keep building the habits of a kind, professional person, even when it is only a chat window. John told me that angle had not crossed his mind.
Then I asked the question I could not drop. Say an employee on a team account gets cut off for abusing Claude. Does HR start to treat that as a red flag?
John said it opens a Pandora’s box. He would call it misuse of a corporate asset. It is not cost neutral, and the argument cannot be won. He could not picture HR treating it as abuse of a coworker. I would not go as far as abuse of a coworker either, but I do wonder. If someone is rude to the AI, are they rude to their team in Slack? If they manage people, how do they manage them? Maybe. I do not know. John pointed out the person could be putting on a persona, polite everywhere and then unleashed where they can be. In some cases it reflects a wider pattern, and in others probably not. We landed on rude instead of abusive, and on not knowing.
One thing I also tried to explain on the air is how a chat works. Inside one conversation, the whole history goes back to the model each time you send a message. The longer it runs, the more context goes in. John added that it can degrade and lose the thread, which I can only imagine makes an angry user angrier.
Owners are buying AI tools, not training people
Forbes had an article about small business and AI. Owners are buying more tools, but the training is not keeping up. I did not catch the numbers, and John said the article he was looking at was behind a paywall, so he only saw part of it. He thought some PwC studies pointed the same way and wanted to check the sources before leaning on them. So treat those numbers as unconfirmed. The direction is what we both agree on.
John has a background in change management and corporate training, so he feels strongly about this. Deploying AI is far more than learning a new tool. A workforce that is less and less trained in responsible use is a risk to your business. The reputational damage and the trust you lose with clients are real. And if your business case depends on productivity gains, but people use the tool for things that have nothing to do with work, you will see usage go up and results stay flat.
I keep coming back to change management. When you upgrade software, someone comes in and shows people the new process. With AI it feels like everyone threw that out. Here is a gun, figure it out. Here is a car, learn to drive on your own. You spent the money on the tool and did not teach anyone to use it.
I still get asked how to prompt it. My son’s answer would be, bro, no. You do not have to think about prompting the way you used to. It is about context. What are you trying to do, and have you given it enough to understand the goal? The models are good at reading between the lines.
The adoption curve matters here too. Hand a tool to everyone and a few people will love it. Most will think, I still have to get my work done, nobody is teaching me, so I will keep doing it the way I always have.
John’s point was that AI is not an ERP rollout. An ERP is a game of widgets. You get an input, you get an output, you get your training manual and off you go. With AI you stumble onto other things it can do, and suddenly you are off in a different direction. He calls it a literacy issue. People need a broad sense of what it does, how an agent works, what context is, what chain of thought is, how to check the output, and what is safe at work. He does not think the employer has to do all of that alone. The person using it has a stake in learning too. And you can use the tool to be part of that learning, so your experimenting helps redesign the work.
His bottom line is that cutting training is counterintuitive. What you save up front you pay back later in a higher burn rate.
Kids, school and AI literacy
I have an eleven year old in sixth grade. He wants to use Claude Code and start building. I talked to his school principal about how they use AI after the teachers handed out some paperwork, and my son said it did not sound like a person wrote it. Kids pick up on that fast. They will grow up with this and probably depend on it. They will also learn to use it.
John said AI literacy should start young. Think about how we teach someone to speak English or another language. It starts early and builds on itself, and the workplace adds its own vocabulary on top. It is a collective effort.
That brought us back to the start. If we teach kids it is okay to be abusive to the AI, that leaks into how they treat people. John called it a nice bookend. I said that if a kid being cut off teaches them a lesson early, good.
He also wondered what will happen with friendly bots marketed to children. That scares me. I grew up around TV, the last twenty years were raised by social media, and the next twenty may be raised by AI, which is far more powerful than either. John wants a global approach to regulation. His comparison was aviation. Without international standards and air traffic control, flying from city to city would be a mess.
The line I left with
I signed off with be kind to the robot overlords, and John said he has nothing but admiration for them. It was a joke. The part I am not joking about is that we ran from rude chats to team training to kids in about half an hour, and it all lands in the same place. People need to be taught how to use this, and they need to build good habits while they do.
Key takeaways
Buying the tool is not adopting AI. Owners are spending more on tools, and training has not kept up. Nobody taught the team.
Cutting training costs more later. John’s view is that what you save up front ends up in a higher burn rate.
Skip the prompt tricks and give it context. Say what you are trying to do and why. The models read between the lines.
AI is not an ERP rollout. People find uses nobody planned, so the training has to be literacy, not a manual.
Your habits with AI are your habits. How you talk to Claude may say something about how you talk to people. We did not agree on whether HR should care.
Literacy starts young. Kids will grow up with this, so the habits and the understanding have to start early.
Timestamps
| Time | Topic |
|---|---|
| 00:00 | Welcome back, John Matoso of SubTonique |
| 01:38 | This week: Anthropic’s new rule on abusing Claude |
| 04:44 | Why AI is built to keep you coming back |
| 06:16 | Does how you talk to AI show up in how you lead? |
| 10:03 | How a chat works: it re-reads the whole conversation |
| 11:52 | Should an employee being rude to Claude be an HR red flag? |
| 15:35 | Owners are buying more AI tools, but not training people |
| 19:55 | Stop asking how to prompt it |
| 21:00 | Handing everyone a tool is not adoption |
| 21:22 | AI is not an ERP rollout |
| 22:05 | AI literacy: what people need to know |
| 24:06 | Cutting training costs more later |
| 24:49 | Kids, school and AI literacy |
| 27:09 | Back to being rude to AI: what we teach kids |
| 28:14 | Marketing AI to children, and global AI rules |
| 31:02 | Sign-off |
About the guest
John Matoso is a change consultant with close to 25 years in communications, training, business consulting and change work. He launched SubTonique about three years ago, around the time ChatGPT came out. He is based in Quebec.
If your team has the tool and no one has taught them to use it, that is the work I do as a fractional CAIO. Here is how that works.