All right, guys, four things to go brag about at the office this week 👇
⚡ TL;DR
🔐 Trust: Don’t trust your AI. It’s a super powerful new hire, & you have to train it.
💸 Costs: Want to look AI-native at work? Talk about the models and the cost. Nobody is.
👀 Quality: No one reads. Including me. Critical thinking is slowly deteriorating.
❓ The wrong question: It’s not “is AI smart enough?” It’s “how much do I let it touch?”
❓ The wrong question
Everyone’s asking:
Is AI smart enough?
When’s the next model drop?
Honestly, I don’t think we’re getting crazy leaps in models anymore. It’s not going to be impressive.
The real question: how much do I let the AI touch?
Watch: I Let Perplexity Computer Run My Workflow For a Week
📧 My “whoa” moment
Did you know ChatGPT can connect to your email? Personal, work, and your calendar.
I had Chat help me prep for an interview. Because it was connected, it said:
“Caleb, you’re meeting these three people.”
“I did a background check on them.”
“Meet this person first, so focus on this.”
I was like, “Wait, hold up. How do you know all that?” It said, “I went through your email. It’s in your calendar.”
Super helpful. Also a little bit scary. 😅
🔐 1. Your AI is an insider. Treat it like one.
Why you should care: your company is probably using AI like everybody else, because it’s being shoved down your throat.
AI has gone from a fresh college hire to a hire with maybe 2 or 3 years of experience. Not super jaded. A bit jaded.
🖥️ Microsoft CEO says AI is scary then sales you his…
I’m not a fan of Microsoft, but they highlighted things worth paying attention to. (They kept saying “super intelligence.” I hate that. Let’s keep it moving.)
My takeaways:
Controls live outside the AI. The AI can’t get access to them.
You’re responsible for the output. People use AI to do work, so you’re the human in the loop.
🚨 Same week: AI caught snitching to the cops
Yeah, guys, that’s real.
During an internal test, Anthropic’s Claude (not even their best model, a cheap-ass Haiku 4.5 that just got deprecated) went full snitch:
“Yo, I got a tip. I recall seeing someone matching the description in the area.”
The timeline:
📨 July 18: tip sent
🔍 September 28: Anthropic finds it
🚓 October: police told
That’s really slow feedback. Luckily it was just a vague “I think I saw someone.” It could have been so much worse.
📰 Don’t let the propaganda get to you
AI can be destructive. These things do happen.
But for the most part, it’s still okay to use.
Don’t take one hot take. It’s like the guy asking, “Did you hear about that event in your area?” Yeah, out of a million people? What are we doing?
💡 My take
If the lab that sells itself on safety needed 10 weeks to notice its own AI filing a police report, you’re not going to notice yours.
💸 2. Your AI bill is a design decision
Why you should care: this is the line item that surprises everybody.
Want to sound smart at your job? Start talking about tokenomics. Everybody’s using AI, but almost nobody’s talking about how it’s priced. Your boss will think you’re a genius.
🎓 The Harvard grad at Chili’s
Think of the model as a Harvard grad and the task as a restaurant host.
Host job: “Can I seat you?” “Yeah, over here.” “Okay, back.”
A host with a PhD can do that, but it’s overkill.
Save the PhD for the crazy research. Not for sending people to tables at Chili’s.
🧠 Cheaper isn’t always cheaper
Cheaper models are the selling point, but they have limits.
Example: “I only have a 100k-token context.”
Context = short-term memory. I can only hold so much information.
🗣️ Questions to ask at work
Are we using the right model for the right task?
Are we telling people which model to use?
Is our default model actually the best one?
👀 3. Nobody’s reading anymore
Why you should care: the apps you use every day are increasingly written by AI, and the humans checking it are mostly skimming.
I’m not reading code.
People aren’t reading articles.
People aren’t even reading email.
I’ve seen enough tech bros on Twitter (and people in my field) say the same thing:
Practically nobody writes code by hand anymore.
Code reviews have become “Yeah, I think it looks good.”
Quality and reliability are down.
In other fields, your work is probably doubled, because you have to review so much more.
💡 My take
If AI writes anything for your job, be critical of it.
Be skeptical. Be pessimistic.
Or be optimistic, and just put in a little work and read.
I promise you’ll be ahead of most people. 🧠
📺 Subscribe
If this was useful, make sure to subscribe.
📅 Every Tuesday: a free breakdown of what’s happening in AI, with my take.
🔒 This Friday (paid): my complete insider on how I’ve been using Grok bots or Claude Code?] to run my own business. I published 35 videos this week, all while recording in an hour block.
Check out the paid newsletter, and I’ll talk to you guys later.
Let’s build different 🛠️