The Barefoot Truth
As my barefoot children, 6 and 3, sprint past me up our hill, their feet reading the ground with primordial intelligence, I’m reminded of Marshall McLuhan’s warning: “Every augmentation is also an amputation.” My children, unlike me, have not yet made the Faustian bargain with society to wear shoes as they travel through their day. For me, the sharp stinging as I walk over sharp gravel reminds me that pampered feet have traded their primal toughness for the seductive embrace of soles and societal norms.
While I can recommend my favorite barefoot shoes (Vivo Barefoot, Earth Runners), what I am talking about is Marshall McLuhan’s notion that “Every augmentation is also an amputation” (Relevant Musings). When you choose to augment yourself, you inevitably must give something up. By embracing shoes, I extended the distance I could travel, opened myself up to specialized shoes like cleats and clip-ins, and made my existence more socially acceptable. For all this, I traded my feet’s hard-won resilience, their ability to read the ground beneath them, their muscular intelligence built over millions of years of evolution. Even the most innocuous tool changes us in visible and invisible ways. The question isn’t whether to make these trades, it’s whether we’re making them consciously, with full awareness of what we’re gaining and what we’re losing.
McLuhan’s Prophecy
Marshall McLuhan was a philosopher who died before the internet and chatbots came to be (1980), who is most known for his work discussing the soon-to-be World Wide Web, which he discussed in his work on the Global Village and the Gutenberg Galaxy. What is likely most relevant today is the notion that every form of automation or augmentation we adopt should also be viewed as an amputation (recently discussed by Neal Stephenson). In other words, as we adopt a system that replaces one of our existing skills, that skill will atrophy out of existence. In many ways, this is the unwritten law of technological progress. We strive to make parts of self better, such that we replace entire skills or organs. Never has this made more sense or moved quicker than it has with the growth of GenAI tooling. GenAI tooling makes it relatively easy to replace one, many, or even all of what one’s professional self may be.
The Hedgehog and the Fox
With automation we trade understanding for control. Never is this more dangerous than when we give up control on things we do not understand. Where would a thought piece about AI be without a mandatory connection to Isaiah Berlin’s age-old insight that “the fox knows many things, but the hedgehog knows one big thing”? The danger isn’t that we become hedgehogs (deep specialists), but that we become false hedgehogs.
Instant access to graduate-level expertise creates the illusion of universal mastery. Even if you are not an expert in financial markets, you can automate that part of your life with AI! Right? Right? Make no mistake, if you do not understand the topic which you claim expertise in, you are not a hedgehog but merely a yes man for your model. If you vibe code your way to a web app that you can’t make heads or tails of, you are not Rick Rubin, you are one hallucination away from a broken build. When everything is pre-chewed for you by an AI system, you never learn to understand what weird looks like and when something isn’t within the operating parameters.
Fire Your Analyst: Mental Seduction in the Age of AI
Automation with AI is centered broadly around this notion of improved profitability by reducing human costs. If we dissect the pitch of hip tools like Claude Code or Rogo AI, the core is to reduce your load by handing over the keys to the AI. In the use of these tools, it’s very important to know the difference between conscious process design and technological sleepwalking. Adopting these new tools that do the jobs of your junior people is very alluring. Save money, do more, etc. Adopt these tools blindly and you won’t know your knowledge gaps until you have a crisis.
A question for all those firing their juniors: when things go bad and you need more people who understand your business and tools deeply, what do you do? Hope and pray to the latent gods to fix things. Good luck. Could these models, in theory, do everything? Yes, Hilbert space huge, but currently, they do not. They lack context of your business (partially because you aren’t good at giving it to them), miss core components, and most importantly, are merely reasoning through space seeking to optimize the reward model, which is likely at odds with your business to some degree. Moreover, one of the things your analyst does for you that you probably don’t respect enough is manage the risk of constant and costly cognitive offboarding. When agents are around, your brain is unlikely to be able to understand their work properly and give them the stamp of approval better than your grandmother could.
Burning the Learning Ladder While We Climb
Bloom’s two sigma is a really interesting way to think about learning because it shows the importance of one-on-one tutoring and the importance of competence-based teaching. Even with all the encyclopedias of the world, if you do not go on that journey of going from novice to expert, you are unlikely to be able to respond well when things are not going well. Build broken at 3 in the morning, and Claude just keeps on trying to hack your tests? Stock down 20% and Hebbia’s matrix telling you everything is fine? Unless you learned how to do these jobs, you will not understand when things fail.
While this can be scary for you, it’s even more important to think about this for the next generation of workers. If you only vibe code your way through code or investment memos, or scientific research, you will not be able to think about the complex interdependence between things in your stack. Understanding how your company’s security policies and existing physical infrastructure impact API design can lead you to avoid huge security holes. If you never let people learn about this, how can we ever expect them to understand what context to provide to models?
Don’t Shun Amputation, Understand It
When you are thinking about improving your productivity, don’t go ahead and think like a Luddite. Computers aren’t evil, and you shouldn’t shun augmenting yourself. When designing augmentations, here are some rules of thumb I have been working on:
Can you verify it? If you are automating something, you need to be able to prove it is working and understand when it breaks. This means you need to be able to verify that the output is what you expect. In coding, this can be found in the use of unit tests and build systems. Think about your build systems and how they can be used to confirm things work as expected.
McLuhan Test: Before you adopt a new tool, map its amputation and make sure it lines up with where you want to be. If an augmentation helps source 10x the content for reading at the expense of my wandering on the internet, I am happy to take that amputation.
Context Collapse: To what extent do you need to be part of the whole story to understand the whole outcome? In some tasks, not watching the first few minutes will lessen your ability to jump in at the last few. Is there a collapse in performance due to a lack of context? If so, be hesitant.
Prioritize Agency: The goal of any augmentation is to give you more agency. Think about whether this task allows you to do more of what you care about or if it makes you better at doing what you care about. Just because a system can read S-1s faster than you doesn’t mean you should give up on stock picking.
Create Feedback Loops: When automating yourself, it’s important to create feedback loops with your augmentation and your other systems. When systems improve efficiency, make sure these gains are seen down the road, and when systems fail, be sure you know how to limit the blast radius.
The Future Belongs to Conscious Architects
The winners of the future will not be those who do the most automation, but those who design their automation in a way that allows them to do more of what they succeed at. It’s not about making generalist agents that do everything, but creating augmentations that make you and your skills a lumpy T. Run toward your augmentation, but make sure to walk barefoot and not run everywhere in marathon shoes. Keep those soles tough.
If this is interesting, drop me a line. I’m new to this writing thing and trying to figure out my voice. It probably has at least some sass to it, but beyond that, I am still working on it. Find me on X.