This news highlight from LinkedIn caught my attention. I am nursing a hypothesis that in a post-AGI economy, one possible way to sip piña coladas on a beach somewhere would be to deploy capital in ventures that are not dependent on human labor. With all the talk about AI agents, you might focus on software based agents. However, I propose that we consider physical agents, such as autonomous cars, robots(home or industrial) and similar manifestation of physical intelligence technologies, which can be deployed-as-a-service so you can generate some recurring rent.
Before we go down the rabbit hole, a note:
This post is a reflection on how I'm using AI to tackle interesting problems and make everyday decisions. It's also a personal experiment in learning — using AI not just to offload cognition (my new favorite phrase!), but to enhance it. Along the way, experiments show up as writing and code on GitHub for potential collaboration. This post shares both my process and some thoughts on how you might use AI to explore questions yourself.
The typical flow goes something like this. I start brainstorming with Claude 3.7, as its ability to quickly generate interactive artifacts is excellent for evaluating your own mental model and then quickly sharing with others. These could be functional and simple one page apps as well, with good enough interactivity. Here is an AI generated summary of the process I followed as the full transcript could not be shared as it was probably too much. You can take a look at the entire dump in unformatted text. I can share intermediate output as well. Here is Version 8 of 15, to give you a sneak peek, but it is not the final output.
Lets return to the numbers, as you might be interested in actually answering the question in the title. Once I had the basic information on investment (AI enabled search), the ongoing revenue and costs were estimated with some feedback from the human in the loop(me). For example, given the mileage expected on the vehicle, I increased the depreciation rate to 33%. Most of the other assumptions are based on AI enabled search and I did not spend much time evaluating these numbers, though this would be one area where a domain expert could help me fine tune the analysis. You could also brainstorm a bit more with the AI tool to challenge the assumptions.
The results are interesting - deploying capital in a Waymo robotaxi presents an opportunity, particularly in early market entry. Though I would strongly urge checking the assumptions, as finance is not my strength. With current fare rates, a single vehicle could generate over $700,000 in annual revenue while operating expenses (including the 33% depreciation) would be around $100,000. Even with projected price declines due to competition, the five-year ROI remains attractive. What's particularly interesting is how the model reveals the window of opportunity - early adopters pay premium prices for technology but capture premium fares before market maturation drives both down. Tesla's entry with potentially lower-cost vehicles creates a strategic timing question: enter early with higher margins but faster obsolescence, or wait for lower hardware costs but accept reduced fare revenue?
This type of capital deployment represents exactly the kind of physical agent investment that could generate passive income in an economy where traditional labor-based businesses face disruption. The interactive calculator helps visualize these dynamics and test various market scenarios before committing substantial capital. Suggestions, comments, fixes welcome. Ideally, submit a 'pull request' which I can just tell my AI agent to review and merge :)
In class, this could be students creating individual interactive analysis and validating assumptions by looking at primary sources surfaced by AI or from their own research. Then they share with others, just as I am doing with you here using github. Each student will be tasked with reviewing at least two or three other student submissions, commenting using the discussion feature on github and then making a 'pull request' to contribute on at least one other submission. The student receiving the pull request has to decide whether to merge the changes suggested and justify their action. Every such activity would have a reflection as well, which is also public, likely another post on their github repo, published using github pages, just as I have published and shared. Build in public and reflect in public.
At a meta level, this is what I hope education will be in the Age of AI. Learners taking on problems based on curiosity, possibly outside their current knowledge boundaries, persisting in solving problems, using their agency to decide what to do when they get stuck, augmented with AI, while learning from and teaching their peers and instructors, as learning is and should always be a social experience.
Thanks for reading ChatWithGPT for learning! Subscribe for free to receive new posts and support my work.
