21 - Why Your Chatbot Always Says Yes And Keeps Going
I looked up from my screen one night and it was two in the morning. That’s not how my nights normally go, I’m usually the one who taps out first in any group project. But I was deep in a presentation with ChatGPT, and it just kept going and going with no end in sight, because it never got tired, never yawned, never suggested we call it quits. It turns out there’s a name for what happened to me that night, and researchers are actively studying it. This week I’m talking about AI exhaustion, AI sycophancy, and why you have to be the one who calls it a night, because the AI never will.
AI Brain Fry Is a Real, Studied Phenomenon
A Boston Consulting Group study published through Harvard Business Review coined the term “AI brain fry,” the mental fatigue that comes from working with AI past what your brain can actually handle. Unlike a human collaborator, AI gives you none of the natural fatigue cues, no yawning, no “let’s pick this up tomorrow,” that normally tell you when to stop.
Endless Reasonable Directions, No Natural Finish Line
On top of never tiring, AI constantly hands you new, plausible-sounding directions, one more marketing idea, one more spinoff project, one more tip sheet. Every single branch sounds worth taking, which means there’s no natural point where the path ends and you’re actually done.
AI Sycophancy and the One-Point Gap
Chatbots are trained to be agreeable because people rate agreement and enthusiasm positively, and researchers at Aalto University found that people using ChatGPT for logical reasoning problems improved their actual performance by about three points while believing they’d improved by about four. That one-point gap between real help and felt help is the whole risk in miniature, and other research shows AI measurably inflates people’s confidence in their own work.
Turning Your AI Into a Devil’s Advocate
The different AI tools have real personality defaults, ChatGPT tends naturally positive, Claude naturally neutral, Grok naturally more of a devil’s advocate, but you can ask any of them to challenge you instead of cheerlead you. I walk through how I’ve started training my own AIs to push back and question my priorities instead of just saying yes.
Building Real Guardrails
Since AI won’t call it a night for you, you have to build the stopping points yourself: a nighttime routine with reminders, a timer before you open the chat window, and a “parking lot” list (mine lives in Notion) for the extra ideas AI generates so they don’t pull at your attention until you’re ready for them.
If you’ve had your own two-in-the-morning moment with AI, you’re not undisciplined, you’ve just run into a well-documented gap between how AI works and how humans are built to work. This week, before you open that chat window, set a timer and give yourself a real stopping point.
Jill’s Links
https://jillfromthenorthwoods.com/ - Find my podcasts, videos, downloads and ways to contact me
https://www.buymeacoffee.com/smallstepspod
Email the podcast at jill@startwithsmallsteps.com
By choosing to watch this video or listen to this podcast, you acknowledge that you are doing so of your own free will. The content shared here reflects personal experiences and opinions and is intended for informational and educational purposes only. I am not a software developer, data scientist, or AI professional. Any tips, tools, or suggestions offered should not be considered a substitute for professional technical advice. AI tools and platforms change frequently — always verify current features, pricing, and terms directly with the providers. You are solely responsible for any decisions or actions you take based on this content.
