This Weird Trick Turns AI into Your Toughest Critic

February 15, 2026

Happy Sunday.

The team and I are busy this week building AI agents to run different areas of our business. Max, Jacob, and I are loving Claude's most recent update.

For our content writing and business strategy needs, the latest version of Claude's "Opus" model has been fantastic.

 

What's Inside Today's AI Newsletter

  • The "Potato Prompt" that forces AI to stop being a 'Yes Man'
  • Claude can now connect to your Apple Health data
  • Claude Code or ChatGPT Codex?
  • This week inside the Smarter AI Community
  • Live Smarter


The "Potato Prompt" That Forces AI to Stop Being a 'Yes Man'

Summary: Here's one most people haven't heard of yet. It's called the "Potato Prompt" and it solves one of the most annoying things about ChatGPT...the fact that it agrees with everything you say. You know what I'm talking about. You share an idea, ChatGPT tells you it's brilliant. You share a terrible idea, ChatGPT tells you that one's brilliant too. The Potato Prompt fixes this with a single code word.

At the start of a new ChatGPT session, you tell it something like: "If I write 'potato' at the beginning of a prompt, be direct. No hedging. No compliments. Find the holes in my logic and tell me what I'm missing." From that point on, whenever you type "potato" before your message, ChatGPT switches from cheerleader mode to critical thinker mode. When you want normal ChatGPT back, just stop using the word. You can also type "unpotato" to explicitly tell it to revert. Simple as that -- one word on, one word off.

The best part? You can save this instruction in ChatGPT's Custom Instructions (Settings > Personalization > Custom Instructions) so it works across every new chat automatically. You don't have to set it up again.

Try this first: Go to ChatGPT, open Settings > Personalization > Custom Instructions, and paste this: "If I write 'potato' at the beginning of a prompt, be completely direct. No hedging, no compliments, no filler. Challenge my thinking and point out what I'm missing. When I stop using the word, return to normal." Then open a new chat, type "potato" followed by a business idea or a piece of writing you've been working on. See what happens.

This is exactly the kind of prompting system we teach inside the Yellow Belt certification -- building reusable shortcuts that make AI work the way YOU think, not the other way around.

Read more: BGR: The Potato Prompt

 

Claude Can Now Connect to Your Apple Health Data

Summary: Back in December, one of our community members asked if there was a way to connect Claude to their Apple Health data. At the time, it wasn't possible. Well...it is now.

Anthropic launched a health integration in January 2026 that lets Claude connect directly to Apple Health (and Android Health Connect). Once you opt in, Claude can summarize your medical history, explain test results in plain language, spot patterns across your fitness and health metrics, and even help you prepare questions for your next doctor's appointment.

This is big for anyone who's ever stared at their Apple Health dashboard and thought "I have no idea what any of this means." Claude can now read all of that data and give you a clear, plain-English summary of what's going on.

Privacy note: You control exactly what data gets shared. You can revoke access at any time. And Anthropic has confirmed that your health data is never used to train their models.

Try this first: Open the Claude app on your iPhone or Android device. Go to Settings and look for the Health integration option. Connect it to Apple Health (or Health Connect on Android). Then ask Claude: "Summarize my health trends from the last 30 days. What should I pay attention to?" See what it finds.

Available to Claude Pro and Max users in the US (beta).

Read more: MacRumors

 

Claude Code vs. ChatGPT Codex: The Battle of the AI Programming Assistants

Summary: If you've been hearing about AI coding assistants and wondering which one is worth your time, here's the breakdown. Claude Code and ChatGPT Codex are the two biggest players right now, and they work very differently.

Claude Code runs in your terminal. It's local-first, meaning it works directly on the files on your computer. You describe what you want, Claude makes a plan, shows you the plan, then does the work. It has features like Plan Mode, memory files that remember your project preferences, and background agents that let you kick off tasks and come back later. It's built for developers who want careful, production-grade code with a developer-in-the-loop workflow.

ChatGPT Codex runs in the cloud. You push your code to it, it works in a sandbox, and you pull the changes back. It also offers a CLI and IDE extensions for VS Code and JetBrains. Codex is fast at rapid prototyping and has strong IDE integration for people who prefer working inside their editor rather than the terminal.

Here's what the benchmarks say: Claude Code writes roughly 1,200 lines in 5 minutes compared to Codex's 200 lines in 10 minutes. But on accuracy, they're neck and neck -- both score around 80% on the SWE-bench coding benchmark. Speed vs. depth. Both are about $20/month at the base tier.

The real answer? A lot of developers are using both. Claude Code for fast implementation and iteration. Codex for code review and quality assurance. They're not really competitors...they're complements.

Try this first: If you're a developer (or aspiring developer), pick the one that fits how you already work. Prefer the terminal? Start with Claude Code. Prefer your IDE? Try ChatGPT Codex.

Read more: Mivi Analysis


Community Highlights

Max Bernstein just dropped something big inside the Smarter AI Community this week -- his Framework Factory Course.

Here's the idea behind it: you've been doing the work for years. Clients refer others to you. But when someone asks "what makes you different?" your answer sounds like everyone else's LinkedIn headline.

Max built the Framework Factory to fix that.

It's a 5-phase system for extracting the frameworks you're already using -- the patterns hiding in your client conversations, the methodology your brain filed under "obvious" -- and turning them into professional, validated intellectual property.

The 5 phases:

  • Recognition (spot the frameworks hiding in your work)
  • Extraction (use AI prompts to mine your transcripts for patterns)
  • Architecture (name it, structure it, design a visual)
  • Validation (prove it works with real evidence)
  • Monetization (build your portfolio and revenue path)

It's built in Notion with interactive exercises, AI prompts you can copy and run on your own content, and a full video walkthrough. No generic templates. You work with YOUR content, YOUR expertise.

If you're a member of our community, go login and check it out. 

If you're not yet...come see what Max and the rest of us are building.

Join the Smarter AI Community today

 

More From Smarter Living

 

Top 3 Tools We're Using Right Now:

Wispr Flow -- Voice-to-text that types wherever your cursor is. Dictate emails, messages, docs...anywhere. I use this every single day.

Tella -- Beautiful screen recordings with no editing required. Record your camera, screen, or both and Tella handles the polish. All of our certification course videos were recorded using Tella.

Granola -- AI meeting notes that actually work. It sits in the background during your meetings and writes structured notes automatically. No more scrambling after calls.

 

We all have that 1 friend...send this newsletter to someone who could use an AI potato in their life.

Missed last week's newsletter? Click here to read it on our website

 

Live Smarter.

Ryan Hutchinson Ryan Hutchinson
Founder, SmarterLiving.ai