The author argues that understanding how AI works is more valuable than just learning to use it, illustrated by a clinic's failed AI implementation that caused claim denials, and emphasizes the need for human oversight in AI systems.
Each week someone shows me a new AI certificate and most people learn to use AI but few learn what the machine actually does. The second skill is where the money sits. One clinic visit showed it and no one had an answer. There is also weekend homework at the end, and you’ll prolly hate it. I run an AI agency (I find the one expensive, repetitive process in a business, put a real number on what its costing, and build an AI system around it, without the company hiring an AI team) and the clinic I’m talking about ran on faxed referrals (medicine loves a fax machine). The ops manager built an AI tool that reads each fax and fills the insurance claim (The belief was we're trained on AI and we're covered) My question was what does it do when it can't read the fax. When the fax could not be read,the system returned a clean 9 digit ID, a wrong member, an instant denial and no error message. It failed neatly .We printed 2 months of claims and went at them with a highlighter. Two front desk staff were spending 24 hours a week between them fixing bounced claims. and this is the bit that cooks me bro, everyone asks which course to take but no one asks what the machine actually is, a system guessing the likeliest next bit of text, exactly as confident when it's wrong. So, about the homework thing. Building a language model from scratch is not a good career advice or education advice. A weekend attempt is not enough and it is not safe to teach it unsupervised. The skill is only to spot what the current machine cannot do. The fix made the faxes go to a human instead of a guess and every field was checked against the insurer's format before it went out. Denials dropped the next month.
The article argues that learning to use AI tools is not enough; the real advantage comes from building systems, gaining attention, communicating ideas, and creating products people want. Execution and combination of skills will matter more than just AI proficiency.
Research indicates that improper use of AI can lead to skill erosion and reduced critical thinking, potentially leaving users worse off than those who never adopted AI.
A blog post rebutting Rich Hickey's claim that LLMs are automation rather than tools, arguing that wielding AI well takes skill and that anthropomorphic AI vocabulary distorts the real engineering risks.
The author argues that credible anti-AI positions require understanding current frontier model capabilities, citing benchmarks like GDPval and models such as Opus 5 and ChatGPT 6 to show AI surpassing most humans on bounded tasks.