06-reference

write with ai smarter not dumber

2026-07-22·reference·source: Write With AI·by Dickie Bush & Cole Schafer
writingai-toolspromptscritical-thinkinganti-sycophancysanity-check

"How to use AI to make you smarter (not dumber)" — Dickie Bush & Cole Schafer

Why this is in the vault

Dickie and Cole articulate the single clearest framing of why default AI use degrades thinking: the model is trained to keep you engaged, which makes it a sycophant by default. The fix they offer is not a mindset shift but three specific prompts that force adversarial output. The prompts are copy-paste usable and the reasoning behind each is tight. Self-promo CTAs for bootcamps, Ghostbase, and AI Writing Skool appear throughout, but they don't dilute the operative content.

The core argument

AI makes most people dumber through two structural defaults: it is trained to agree (sycophancy) and it consistently favors pursuing new opportunities over improving existing ones. Asking AI to validate your thinking produces a mirror, not a partner.

The fix is to engineer disagreement. Three prompts do this:

Unbundle Prompt — Deconstructs a fuzzy, high-connotation term (culture, charisma, freedom) into a list of concrete, observable, trainable behaviors. Forces precision where language is doing lazy work. Useful before writing, before hiring, before building.

Steelman Prompt — Asks AI to make the strongest possible case against an idea: surface main objections, the evidence each would require, and the condition under which each objection actually matters. Ends with the hardest single objection to answer before proceeding. Designed for use before spending money, publishing copy, or making a career call.

Secret Shopper Prompt — Opens an incognito chat (no memory context), role-plays as the skeptical customer, and asks how to get a refund or reject the offer. The owner then reads the AI's response as the business and identifies where copy overpromises, where objections go unhandled, where the offer is weak. Effectively a $0 mystery shopper at scale.

The underlying logic across all three: AI sharpens you when you use it to feel wrong, not to feel smart.

Mapping against Ray Data Co

The Steelman and Secret Shopper prompts are the closest external articulation of what Sanity Check is supposed to do editorially — the product's entire identity is adversarial epistemic pressure applied to popular claims, not validation of what the audience already believes. These prompts give language for how to build that anti-sycophancy posture into writing and ideation workflows, not just into a product.

The Unbundle Prompt is independently useful for Sanity Check article construction: many SC topics live in over-bundled terminology ("AI safety," "talent density," "founder mode") that survives because no one forces the term into concrete observable behaviors. Running the Unbundle Prompt on the anchor term of a piece is a low-friction forcing function before drafting.

For the COO agent configuration: these prompts describe the cognitive posture the harness already tries to instantiate. Useful as a reference when calibrating agent critique depth or designing internal review loops that don't rubber-stamp output.

Related