Caio and Marina face an uncomfortable truth nobody who uses AI every day likes to admit: you can get more productive and less capable at the same time. Research from Microsoft, MIT and Anthropic, real examples from code, reports and manager decisions, plus a practical test to find out which group you're falling into.
In this episode
01 The hook: productive and dumb at the same time
- Caio opens with a confession: he uses AI every day, builds systems with AI, and that's exactly why he started to worry. Marina pushes back on the paradox: 'Wait, you make a living off AI and you're badmouthing it?'
- The line that anchors the episode: for the first time, a technology makes people more productive and less capable at the same time. Marina asks if that's not a bit dramatic.
- Caio pulls the Microsoft–Carnegie Mellon data: 319 professionals who use AI daily, and the more they trusted the tool, the less critical thinking they applied. The more you trust, the less you question.
02 Cognitive offloading: the bill we've already been paying
- Caio explains the term without jargon: shifting the mental effort outside your head. Marina translates it with a daily-life example, like not knowing anyone's phone number by heart anymore after smartphones.
- The staircase: first the calculator, then GPS, now we're outsourcing reasoning, analysis and decision-making. Marina asks: 'But is outsourcing mental math the same as outsourcing thinking?' Caio draws the line.
- The four concrete signs: people writing code they can't explain, articles they can't defend, reports they can't interpret, strategies they can't execute. The deliverable exists, the understanding doesn't always.
03 The coders' case and what MIT discovered
- The before and after of the dev: they'd make mistakes, research, struggle, learn. Now they ask Claude, copy, paste, it works. The provocation: if you can't explain the code, do you actually know how to program?
- Marina defends the other side: 'But nobody memorizes everything, we always checked Stack Overflow.' Caio partly agrees and shows where the line shifts when you never process the answer.
- MIT study: people who constantly relied on AI had lower engagement, lower retention and a reduced ability to remember and explain their own work afterward.
- The strongest twist: in a four-week experiment, people using AI to spot false information started off more accurate, but lost the ability to validate on their own over time. AI helped while the person unlearned the process.
04 The manager who became a relay
- Caio points to the most dangerous case: a leader making decisions based on an AI answer they never validated. Doesn't look at the data, doesn't check the source, doesn't challenge the conclusion, just goes with it.
- The line that closes the block: at that point the manager stops being a decision-maker and becomes a relay for answers. Marina asks how you notice you've fallen into it without realizing.
- The productivity paradox: someone delivers in an afternoon what used to take a week, but how much of that knowledge actually stuck in their head? Marina reacts: 'Exactly, I produce way more and remember way less.'
05 The counterpoint almost nobody mentions
- Caio balances it with Anthropic's data: most AI use is still human amplification, not replacement. Around 57% of uses serve to boost a person's capability, not to automate everything.
- The conclusion that flips the script: AI isn't the problem, the problem is how it's used. Marina asks what separates one use from the other in practice.
- Caio's theory on the market's next divide: it won't be between who uses and who doesn't use AI. It'll be between who uses it to learn faster and who uses it to think less. One group gets exponentially better, the other gets dangerously dependent.
06 The practical test: can you still think on your own?
- Caio closes without preaching: AI maybe isn't creating useless professionals, but it is creating people who look competent without necessarily being it — and that difference will get harder and harder to spot.
- The question that ties you in knots: when AI isn't available, can you still think? Marina admits the question bothers even the two of them.
- Three concrete habits to get out of Group 2: before accepting the answer, try to predict it; ask the AI to show its reasoning and check if you agree; and once a week do a task from scratch, no help, just to measure what you still know by heart.
- Caio wraps with the tagline: productivity without understanding is technical debt you only notice when the tool goes down. Use AI to raise the ceiling, not to forget the floor.