With the advent of AI, machines in general, and maybe in the future humanoid robots, why is human scale still needed? It’s because, until these automations are good enough to replace the jobs they intend to replicate, human scale is needed in many things that automation is not good at but have a required ratio of human to needs for good quality and satisfactory quantity.

To fill in potholes across a city before the next winter, you need enough humans to go around and fill them. To treat enough mentally unwell patients, you need enough personnel to treat them where usually it’s a very low ratio of treaters to those needing treatment. To have the most effective learning environment, a lower ratio of teacher to student is helpful for more one on one, personalized guidance. It’s because of costs that quality and efficiency of each of these tasks aren’t the best.

I believe we should be making specialized robots, specialized AI, specialized anything to maximize effectiveness of automation. Making generalized humanoids or generalized LLMs even are inefficient since you’re allocating more resources to cover too many cases that aren’t required in many domains. Specialization is always cheaper and more effective; you simply need to make as many specialized robots as possible. Prior to the talk of robots replacing humans in everything, people have been tinkering on how to make specialized things in general forever. For example, most SaaS websites use one of two relational databases; many databases come with full text search, but, due to the design of these relational databases, they could never implement a fast enough and good enough quality text search database, so programmers invented dedicated databases for searching. Specialization grants efficiency and better quality for specific tasks, and we make specialized things for tasks that are huge factors for success.

Making specialized automated “machines” would be cheaper and better than humans and their training and costs. If we can make automations that are cheaper than humans, machines’ production times and scale up potential are much faster and likely cheaper through economies of scale than humans which require education and incentives.

For now, many tasks simply require a human and may never be able to be replaced by robots. Some things just need a human touch whether that be emotional or connection.

The Physical World’s Future of Scale: Manufacturing

The most obvious example of human scale was the automated transition of farming. Farming output was always driven by the number of working hands. Today, it’s literally driven by human operated machines. Yet, we still require a certain, now smaller, scale of farmers today that operate the machines. Machines lowered the number of farmers, but there still needs to be enough farmers not only to drive them but to own enough land and to manage the rest of the farming tasks like selling, negotiating, etc.

The future is automating how to create specialized robots. I suspect Jeff Bezos’s new manufacturing PE rollup is about automating the creation of robots rather than combining similar manufacturing firms and sharing machines.

Any firm can decide to produce any product they want, but they have to dedicate resources to it. In a B2B SaaS, that means putting engineers, PMs, designers, and sales to work. They’re looking to maximize profit while also reducing the burden of having to maintain and integrate the code later on. The venture has to be worthwhile in the product’s entire lifetime.

I have an acquaintance that runs a Chinese manufacturing and e-commerce firm. They only want to choose a physical product that’ll maximize profits. In their case, their labor is not really just the manufacturing itself; it’s the design of the machines to build the products. For carmakers, they want to maximize the efficiency of the assembly and manufacturing of cars and their parts. In both cases, they made processes as repetitive as possible to maximize the speed of each individual part. In the e-commerce firm’s case, they have process engineers who build individual machines for manufacturing each individual part of their physical product. E.g. if they’re building a chair, they need a machine just to churn out cushions, another to churn out chair legs, another to churn out chair seats. In my acquaintance’s case, their firm’s costs is not just from COGS, supply chain contracts (i.e. must order a certain volume), and marketing, but also the upfront labor costs of designing machines to meet the demand scale.

My bet is Jeff Bezos’s PE firm is about automating the creation of those machines. Manufacturing is all about maximizing efficiency of each individual process or the puzzle altogether. Every manufacturing firm builds out machines tailored to a product, and every product is going to be completely different from firm to firm. Unlike accounting which have extremely repetitive processes in the experience itself, the repetitive work of manufacturing is from the machines themselves, but the machines are totally different everywhere. Thus, the only thing you can automate from a PE/startup perspective is the labor in creating those machines.

I love thinking about the physical world because software is extremely scalable whereas the physical world is not. It takes a ton of time to do a lot of physical world tasks. Building products for the physical world is infinitely harder and more time-consuming than software due to the laws of physics. But anything in the physical world from selling groceries to billions of humans meaning there needs to be grocery stores and enough food nearby to fulfill them to filling in potholes across a city simply requires physical and human scale.