Reader's feedback on "The Great Abstraction"

A couple of days ago, I wrote a piece titled “The Great Abstraction?”, highlighting the impending tension between OS-layer agent and the entire corpus of companies operating in the app layer. Catapult Capital had some constructive thoughts on that piece which I think is worth highlighting. So, I am sharing their thoughts below:

I have been thinking about this as well with respect to Amazon in particular. There are a lot of unknowns still but I think it is an uphill battle to achieve this, even for Google within Android.

A few of the challenges for Google I have been pondering:

1) The biggest underlying challenge is that there really isn’t much of a problem to fix. Amazon works REALLY well already. They show you a broad range of options with best-in-class review depth and an extremely simple and easy purchase workflow. Google will have to stretch to show an advantage big enough to change user habit. The most obvious path is to pitch that they are able to show offerings across the whole internet and not just Amazon, but can they actually achieve that breadth? That brings me to the second challenge.

2) Will Google even be able to include Amazon in this agentic workflow? Amazon can easily ban this in their terms of service and a large corporation like Google won’t subvert that. I suspect the ability to ban agents will be something that is eventually tested in court/congress and I wouldn’t be surprised if Google wins out eventually (i.e. I wouldn’t be surprised if Amazon is forced to allow agents to interact with their website on behalf of users) but this is far from certain and is the first battle Google must win. If Amazon is not in the offering the service will only take share on the margin in my view. Amazon is half of ecommerce and its the important half and the majority of Americans are locked in via Prime. If these users are going to only use one protocol I strongly suspect most will opt for the Amazon app and not the Gemini workflow if it lacks Amazon access.

3) Lets assume Google is allowed to take action on Amazon’s website. The next legal question will be how much of Amazon’s data can Google take and reproduce in its workflow. When I search for sneakers or whatever, what I really want to see is all the different options laid out in a grid the way Amazon does things. For most things I don’t want Gemini to grill me with a bunch of questions and then present a short-list. I want to see what’s out there. So to match this, Gemini will have to show something like the Amazon format, but with the hook being that other vendors are included. That sounds appealing, but as a user, I will also want to see Amazon reviews. I will want to be able to click the items and see more pictures. Maybe read some of the reviews. See the product listing page. Google will have to be allowed to read all this data from Amazon (and other websites) and present it all in its workflow to match Amazon. I suspect this will be a bridge too far for courts/congress and I don’t think they will be allowed to do this. Especially if a product is only on Amazon and its just a straight lift of data from one page rather than a mixing of data across many pages into an aggregate. If Google can’t take all this data, it will be at a serious disadvantage versus Amazon for most purchase workflows.

4) The final thing worth mentioning is the AI capabilities needed for Google to even pull this off without errors and with the lightning quick load times, they will need to compete with Amazon’s offering. It is popular to talk about the “jagged” frontier of AI capabilities and I would note that computer use is one of the areas where AI models are surprisingly weak relative to other capabilities. Think about ChatGPT’s operator system. Models just aren’t completely robust at reading web pages and taking actions over long contexts. Most likely this will get really good at some point in the future but that is another thing that will take time. And then there is also the “long march of 9s” to consider. This has to be done fast and on a reasonable compute budget, but I suspect even 99% success will be too low when the traditional Amazon workflow has 100% success. You can’t order the wrong thing 1 out of 100 times or even 1 out of 100,000 times or courts will side with Amazon and you won’t be allowed to operate on their site. With AI it seems to take just as much effort/time to go from 90 to 99% success as to go from 99 to 99.9% success and 99.9 to 99.99% success. Look at driverless cars and how long that has taken relative to expectations (and how we still have one remote operator for every 5 cars for Waymo to handle tail events). I suspect this will eventually be technologically feasible but it may take much longer than you’d guess.

A final point. The main assumed advantage Google is offering consumers here is offering the whole internet vs just Amazon. It’s worth keeping in mind that Amazon seems to be working hard to close this potential advantage by offering more and more links to outside websites in its search results.

In addition to “Daily Dose” (yes, DAILY) like this, MBI Deep Dives publishes one Deep Dive on a publicly listed company every month. You can find all the 65 Deep Dives here.

Current Portfolio:

Please note that these are NOT my recommendation to buy/sell these securities, but just disclosure from my end so that you can assess potential biases that I may have because of my own personal portfolio holdings. Always consider my write-up my personal investing journal and never forget my objectives, risk tolerance, and constraints may have no resemblance to yours.

My current portfolio is disclosed below: