Latest / Career Skills with Fexingo: Hard Skills, Soft Skills, and Future-Proofing Your Career / How to Use AI Tools Without Losing Your Analytical Edge
Transcript
- Lucas: There is this quietly terrifying number from a Microsoft study published last fall. They tracked knowledge workers who started using AI copilots for complex analytical tasks — data synthesis, report drafting, scenario modeling — and found that after six months of regular use, performance on independent reasoning tests dropped by about 12 percent. Luna: Twelve percent in six months? That is not a subtle shift. That is the kind of decline you'd normally associate with skill atrophy over years. Lucas: Exactly. And the study is careful to say this isn't about people getting dumber. It's about what the researchers call 'cognitive offloading' becoming a default reflex. When you know the AI can do the heavy lifting, your brain stops practicing the muscle of working through a problem from first principles. Luna: So the thing that makes you more productive in the short term — faster output, fewer hours staring at a blank page — might be eroding the very skill that makes you valuable in the long term. That is a brutal trade-off. Lucas: It is. And I think the temptation is to respond with a blanket rule — 'just don't use AI for hard stuff.' But that's naive. These tools are embedded in every major platform now. Microsoft Copilot, Google's Gemini in Workspace, Anthropic's Claude for coding. You can't opt out without opting out of modern knowledge work. Luna: So what is the alternative? Because I don't think the answer is 'use it less.' That feels like telling someone to eat fewer calories without giving them a diet plan. Lucas: Right. You need a protocol. There is a method I picked up from talking to someone who used to work in the research division at Bridgewater. They call it the 'Three-Pass Method' — skim, draft, critique. Pass one: you skim the AI's output to get the lay of the land, but you don't read it carefully yet. Pass two: you close the AI window and draft your own version of the analysis from memory and logic alone. Only then, pass three: you open the AI output again and compare — what did it catch that you missed? What did you see that it didn't? Luna: That is interesting because it forces you to generate your own hypothesis before you get contaminated by the AI's framing. I think that's the key move — the sequence matters. Lucas: Precisely. The contamination is real. There is a well-documented anchoring bias where if you see a number or a conclusion first, your brain struggles to consider alternatives. The Three-Pass Method builds a firewall around your own thinking. Luna: It reminds me of something from Daniel Kahneman's work — 'think slow' before you 'think fast.' The AI is basically a fast-thinking machine. If you let it go first, you never engage your slow system. Lucas: That is exactly the cognitive science underneath it. And the Bridgewater person told me they actually enforce this in certain meetings. If someone brings an ai generated analysis, they have to present their own take first, then reveal the AI's version. It's a cultural rule, not just a personal habit. Luna: That's a fascinating institutional practice. But for an individual listener who doesn't have a Bridgewater culture at their company, what is the one habit they can start tomorrow? Lucas: Before you ask any AI tool for an answer, write down your own hypothesis. Even if it's wrong. Even if it's just a sentence. The act of committing to a position before seeking input changes how you process the input. You become a critic, not a sponge. Luna: I love that. And I want to push on something else. The Microsoft study also found that novice users — people new to a domain — actually improved their reasoning when using AI because the tool helped them grasp fundamentals faster. So maybe the risk is concentrated in experts, not beginners. Lucas: That is a crucial nuance. If you're a junior analyst learning to model cash flows, an AI that shows you the steps can accelerate your learning curve. The danger zone is the experienced professional who delegates the whole process and stops practicing judgment. Luna: So the prescription depends on where you are in your career. A new hire should lean in. A ten-year veteran should lean back and use the Three-Pass Method. Lucas: I think that's right. And it maps to something the venture capitalist Paul Graham wrote years ago — that the most dangerous AI users are the ones who trust it the most. Novices are naturally skeptical because they're still testing the tool. Experts can get complacent. Luna: Which brings us to the question of company policy. Should firms mandate 'ai free hours' or 'ai free tasks' to protect analytical skills? Lucas: A few are trying it. I've heard of teams at Shopify and one of the big consulting firms that designate Tuesday mornings as 'deep work' — no AI tools, no chat, just individual reasoning. The results are mixed. Productivity dips, but the quality of strategic thinking reportedly improves. Luna: But that only works if leadership models it. If the CEO is sending ai generated memos, the junior staff will follow suit. Lucas: Right. Culture eats strategy. And I think the deeper insight here is that the skill we really need to cultivate is metacognition — thinking about your own thinking. Knowing when you're using AI as a tool versus using it as a crutch requires a constant internal audit. Luna: That is a hard sell in a culture that rewards speed. No one gets a bonus for 'demonstrated independent reasoning.' They get a bonus for closing tickets or delivering reports fast. Lucas: That is the tension at the heart of this entire episode. The incentives are misaligned. And I think that is why this conversation matters — because individual career resilience might depend on swimming against the current. Luna: Speaking of career resilience, I want to mention something quickly. We keep this show ad-free by design — no sponsors, no commercial breaks, no product placements. If you find these conversations useful for your own thinking about your career, and you want to support that choice, the link is buy me a coffee dot com slash fexingo. It is a small way to keep the show independent and focused on the substance. Lucas: Yeah, and we mean that sincerely. Listener support is what allows us to spend time on nuanced topics like this one instead of chasing trending headlines. So if you've gotten value, that link is there. No pressure. Luna: Alright, back to it. So we've established that the risk is real, the method exists, and the habit of hypothesis-first is actionable. But I want to zoom out. Is there a broader skill here — something like 'AI literacy' — that goes beyond just protecting your analytical edge? Lucas: Absolutely. I think the meta-skill of the next decade is going to be what some researchers call 'symbiotic reasoning' — the ability to fluidly move between your own cognition and machine cognition, knowing when to lead and when to follow. It's not just about avoiding atrophy. It's about orchestrating a partnership. Luna: That sounds like a skill you'd need to deliberately build, not something you'd pick up incidentally. Lucas: Right. And I think the most practical way to build it is to treat every AI interaction as a two-phase process: first, you do the work, then you use the AI to stress-test it. That flips the default from 'AI drafts, I edit' to 'I draft, AI audits.' That one switch changes everything. Luna: And it preserves your ownership of the output. If you draft first, the final product is yours. If the AI drafts first, you're just a curator. Lucas: Exactly. And curators are replaceable. Thinkers are not. That, in a nutshell, is why this episode exists.