AI Is Making You Look Smarter. Yet It May Actually Be Making You Dumber.
The Performance Illusion
Your emails are better. Your reports are faster. Your ideas sound sharper. Your research takes minutes instead of hours.
You look smarter.
But how much of that intelligence still belongs to you when the AI disappears?
A controlled 2026 study found that people who leaned more on inexpensive, easy AI assistance performed better while they had it, but those who relied on it more performed worse later without it. Their AI-assisted performance also made their underlying ability look better than it really was.
AI can hide reduced independent capability behind improved assisted performance.
We Are Confusing Performance With Capability
This is the central distinction.
Performance: What you can produce with AI beside you.
Capability: What you can understand, judge, challenge, and execute when the AI is wrong or unavailable.
Those are not the same thing.
AI can raise your performance ceiling while quietly lowering your independent floor.
For corporations, this means a company can become dramatically more productive while simultaneously becoming more dependent on infrastructure its employees increasingly cannot audit. This is not merely an education story. It is a CEO problem.
Cognitive Offloading Is Not New
Humans have always outsourced cognition. Writing outsourced memory. Calculators outsourced arithmetic. GPS outsourced navigation. Google outsourced retrieval. And civilization did not collapse.
So the warning is not: Never offload thinking.
The real question is: Which thinking can you safely outsource, and which thinking must remain exercised because it determines whether you can recognize when the machine is wrong?
The Dangerous Shift in Thinking
The danger begins when we offload the very cognitive functions needed to judge whether the AI is right. These include problem framing, judgment, contradiction detection, evaluating evidence, deciding what matters, distinguishing confidence from truth, generating alternatives, and challenging assumptions.
Microsoft's study of 319 knowledge workers and 936 real-world AI-use examples found that greater confidence in GenAI correlated with less critical-thinking effort. It also found that AI shifts human work away from solving tasks directly toward verifying and integrating AI outputs.
The danger begins when verification becomes something you are no longer qualified to perform.
The Junior Employee Problem
How did senior professionals become senior? By doing bad first drafts, basic research, tedious analysis, junior coding, rough calculations, making mistakes, revisions, and pattern recognition.
AI is increasingly removing exactly those apprenticeship tasks.
If AI does the junior work, where do tomorrow's senior people come from?
Efficiency can destroy apprenticeship.
How You Use AI Matters
AI is not inherently cognitive poison. The design of the interaction determines whether it substitutes for thinking or scaffolds it.
A 2026 systematic review synthesizing 67 empirical studies found that ChatGPT can improve critical and creative thinking when used inside structured, inquiry-oriented tasks requiring comparison, justification, reflection, and revision. But unstructured use often produced cognitive offloading and superficial engagement.
Another 2026 systematic review similarly found benefits when AI was integrated into structured learning, while evidence for cultivating deeper critical-thinking dispositions remained inconsistent.
Two Kinds of AI Assistance
A recent psychology paper makes a useful distinction between dependent cognitive offloading and autonomous cognitive offloading. The researchers distinguish between dependent offloading, where users delegate core thinking to AI, and autonomous offloading, where AI is used as a scaffold while the user retains cognitive agency. Their results suggest these modes are associated with very different downstream outcomes.
Dependent AI: "Give me the answer."
Cognitive leverage: "Show me what I'm missing."
That is the difference.
Another contrast:
Assistant AI: removes friction.
Thinking infrastructure: introduces the right friction.
Convenience Is Not Always Your Friend
We have trained the market to believe: fewer steps = better product. But cognition does not always work that way. Some friction is productive.
You want friction when: assumptions conflict, evidence is weak, consequences are irreversible, your confidence exceeds the data, or yesterday's belief contradicts today's decision.
A good calculator removes friction. A good thinking system sometimes creates it.
Recursion Can Either Weaken You or Strengthen You
Bad loop:
- Human asks AI.
- AI answers.
- Human accepts.
- Human asks again.
- AI carries more of the load.
- Human carries less.
That is recursive dependency.
Better loop:
- Human forms position.
- AI challenges it.
- Human revises.
- AI compares the revision against history and evidence.
- Human becomes harder to fool.
That is recursive cognitive development.
Recursion that removes thought creates dependency. Recursion that increases pressure creates capability.
The Recursum Difference
Recursum can give you answers. That is not the interesting layer.
The deeper layer is designed around keeping the human cognitively active: continuity, contradiction, historical comparison, competing perspectives, reasoning pressure, and governability.
The purpose is not to make the AI think instead of you.
It is to make it increasingly difficult for you to think badly.
The intelligence should compound on both sides of the relationship.
Most AI is optimized to reduce cognitive effort. Recursum's deeper layer is designed to increase cognitive quality.
The CEO Warning
Ask the uncomfortable question: If I removed AI from my organization tomorrow, what capabilities would remain?
Not productivity. Capability.
Could employees still: frame ambiguous problems? identify flawed assumptions? challenge an AI answer? operate when systems fail? train the next generation? know when the machine is confidently wrong?
If your people can operate AI but cannot operate without it, you may not have augmented the workforce. You may have hollowed it out.
The Human Test
After using AI, ask: Do I understand this better than I did before?
Not: Did I finish faster?
That one question encapsulates the whole article.
If AI repeatedly produces the work but your understanding does not increase, you are renting intelligence.
And rented intelligence disappears when the meter stops.
Ending
AI should make you faster. It should make you more capable. It should expose what you cannot see. It should challenge weak reasoning. It should help you accumulate better judgment over time.
But if the AI keeps getting better while you become less able to function without it, something has gone wrong.
You did not augment your intelligence.
You outsourced it.
The smartest AI relationship is not the one where the machine has all the answers. It is the one that leaves the human harder to fool.
TL;DR: AI can make your output look smarter while reducing the amount of thinking you actually do. A 2026 study found that heavy AI reliance improved short-term performance but led to worse performance later without AI, as AI can hide reduced independent capability behind improved assisted performance. This creates a critical distinction between performance (what you produce with AI) and capability (what you can do independently). While cognitive offloading isn't new, the danger begins when we offload the very cognitive functions needed to judge whether the AI is right. This also removes apprenticeship tasks, threatening future expertise. Research shows AI can enhance critical thinking if used in structured, inquiry-oriented ways, but unstructured use leads to superficial engagement. The key is distinguishing between dependent (delegating core thinking) and autonomous (scaffolding) cognitive offloading.
Recursion can either create dependency (AI carries more, human less) or cognitive development (AI challenges, human revises). Recursum's deeper layer is designed to increase cognitive quality, making it difficult for users to think badly. CEOs must ask what capabilities remain if AI is removed, warning against "hollowing out" the workforce. Ultimately, if AI repeatedly produces work but your understanding doesn't increase, you're renting intelligence that disappears when the meter stops. The smartest AI relationship leaves the human harder to fool.
By Ernesto Verdugo, AI Architect, Founder of Verdugo Labs, and Creator of Recursum. He builds systems for what happens when human and artificial intelligence stop working separately.
Links: