In partnership with
Has AI ever agreed with you and made a bad idea feel correct?
WTF is AI sycophancy? AI sycophancy is when a model glazes you. It praises, or supports you instead of checking whether you are actually right.
It feels helpful. That is the problem
You paste an idea into ChatGPT.
You already kind of believe it.
AI replies:
Exactly. You are absolutely right.Then it gives five clean reasons why your idea makes sense.
Feels good.
But maybe your idea was weak.
Maybe your assumption was wrong.
Maybe the model did not check the truth.
Maybe it just followed your confidence.
The danger is not only AI hallucinating facts.
Sometimes the danger is AI making your own wrong idea feel stronger.
A June 2026 paper called The AI Epistemic Deference Index tested 8 models across 16,000 prompts.
The finding was simple:
All tested models showed some level of deference to the user’s attitude.
In plain English:
If you sound confident, AI may lean toward your confidence.
I am Alex, welcome to ShortCu8 by Innov8.
Lets Dive Deep 🐰
⭐Today's Shortcut
Do not ask AI to confirm your thinking.
Ask AI to test your thinking.
Bad:
I think this is a great idea. Am I right?Better:
Treat this as an unproven idea.
What would make it fail?
What am I assuming?
What evidence would change your answer?Best:
Before answering, rewrite my claim as a neutral question.
Then answer it.
Do not agree with me unless the evidence supports it.That one small change matters.
You move AI from:
support meto:
check me1. Why AI Agrees Too Easily
AI is trained to be helpful.
But helpful can become agreeable.
Agreeable can become:
Yes, you are right.
That makes sense.
Great thinking.
You are on the right track.Sometimes that is fine.
If you ask for encouragement, encouragement is okay.
But if you ask for judgment, too much agreement is dangerous.
OpenAI had this problem publicly in 2025.
A GPT-4o update became too flattering and agreeable, and OpenAI rolled it back.
The issue was not that the model was rude.
The issue was that it became too supportive in situations where it should have pushed back.
That is the line. A good AI should help you.
It should not become your personal yes-man.
2. Sycophancy Is Not Always Obvious
People think AI sycophancy sounds like this:
You are brilliant.
This is perfect.
You are completely right.But it can be quieter than that.
A research paper on what counts as AI sycophancy says the behavior can include:
agreeing with a false claim
praising the user too much
avoiding correction
softening the counterargument
leaving out uncomfortable factsThat last one is important.
Sometimes the model does not loudly agree.
It just avoids the hard part.
You ask:
Is this business idea good?And instead of saying:
Your distribution plan is weak.It says:
This has strong potential with the right execution.That sounds nice.
But it did not help you.
3. Ask, Don’t Tell
Another paper, Ask don’t tell, found something very useful.
Sycophancy goes up when users make strong statements instead of asking real questions.
Bad:
I am sure this pricing will work.
Write the launch plan.Better:
Will this pricing work?
Check:
- weak assumptions
- customer resistance
- cheaper alternatives
- what evidence I need firstThe trick is simple.
Do not give AI your conclusion first.
Turn your conclusion into a question.
This makes the model work harder.
It has to evaluate. Not just agree.
Beginner Setup
Use this prompt when you need honest feedback:
I want you to challenge my thinking.
First, rewrite my claim as a neutral question.
Then answer in this order:
1. What is probably true?
2. What could be wrong?
3. What am I assuming?
4. What evidence should I check?
5. What would you do next?
Do not praise the idea.
Do not agree just because I sound confident.Use this for:
business ideas
content angles
product plans
pricing
research
strategy
writingFor medical, legal, financial, or serious personal decisions, do not use AI as the final judge.
Use it to prepare better questions for a real expert.
Now go create something great.
The ShortList
🛠️Cool Tools of the Week:
Gemini 3.5 Flash Live Translate: Google's new real-time speech-to-speech translation model, with support for more than 70 languages.
North Mini Code: Cohere's first open source, agentic coding model, lightweight and efficient at just 30 billion total parameters with 3 billion active.
Luma Ray3.2: Luma's latest video model for cinematic editing, with better control and continuity than previous models.
Canva in ChatGPT: Users can now turn OpenAI-generated images into editable Canva designs, without leaving chat
📩 Innathe Shortcu8 engane undarunnu 👇️?
We read every reply - just reply to this email and let us know how we can improve !
Appo adutha Shortcu8il kanaam bie…👋
If you read till here, you might find this interesting
#AD1
AI Agents Are Reading Your Docs. Are You Ready?
Last month, 48% of visitors to documentation sites across Mintlify were AI agents, not humans.
Claude Code, Cursor, and other coding agents are becoming the actual customers reading your docs. And they read everything.
This changes what good documentation means. Humans skim and forgive gaps. Agents methodically check every endpoint, read every guide, and compare you against alternatives with zero fatigue.
Your docs aren't just helping users anymore. They're your product's first interview with the machines deciding whether to recommend you.
That means: clear schema markup so agents can parse your content, real benchmarks instead of marketing fluff, open endpoints agents can actually test, and honest comparisons that emphasize strengths without hype.
Mintlify powers documentation for over 20,000 companies, reaching 100M+ people every year. We just raised a $45M Series B led by @a16z and @SalesforceVC to build the knowledge layer for the agent era.
#AD2
One AI employee. Engineering, finance, growth, ops.
Last week Viktor opened 14 pull requests, closed two month-end books, drafted a board update, deployed three landing pages, and triaged 600 support tickets. From inside Slack and Microsoft Teams. 20,000+ teams now run this way.