AI Shopping Agents

The main benefit of AI in commerce is quickly removing friction by broadening out the scope of search and showing alternate solutions. There is a trade off in any broadening. New ideas and solutions can be valuable, but broaden things too much and create a sea of noise.

Distribution platforms which gained power through removing friction may prefer a bit of extra friction in their offerings to boost monetization in an area they absolutely dominate. Added friction can be things like heavy load on the Amazon search result page, trying to force the brands to re-buy their branded terms to block out competitors, requiring extra clicks to reveal information like phone numbers, and hiding some features like search by image inside the app to try to drive app downloads.

The other day I was walking around with my vintage Nike basketball shorts and did not know they had a hole in the right pocket. I somehow lost the rental car key, then came back & when a second key fell through that hole I realized that is what must have happened to the first key. So then I had to get a new key after failing to find it on about 8 passes through they area I was walking.

Google Shopping is far from ideal in terms of which items get shown when there are many options for goods like clothing. You can use the sidebar filters or put the item product size in your search query, then when you click around the Google Shopping results you keep clicking into landing pages for medium or small shorts, because the incidental incremental & irrelevant clicks make Google more money instead of showing just what you want - similar to the Amazon search results trying to force brands to bid on their own branded terms by featuring ads for knock off goods high in the results.

If you search Google for locksmiths they might sometimes use the AI to give rather specific details about the security specs and process needed to rekey certain car key types, but locksmiths is one of the dirtiest and spammiest categories online, where many players say they can do everything to get your info & they are actually just selling leads off to other service providers.

Every single platform has some sorts of optimizations that are aligned with the consumer (say fast shipping, cheap shipping, saved info profiles on accounts, order history, and broad selection) and some that are aligned against the consumer (false promises of delivery speed, gray market or counterfeit goods, lead reselling that leads to a deluge of calls, etc.)

One of the locksmiths who called me frequently used "my friend" while talking - perhaps one of the more bright red flashing warning lights you can get. Then he comes out quoting a fake dealership key price of $3,000 and then suggests he can come in at just over half that - which is actually hundreds above what the dealership would charge!

As free blogs wither away & the 5 or 6 big web channels keep filling up with re-represented content from each other you'd think that sites like Patreon should be doing strong as people notice the difference between AI junk and creators they are real fans of - yet the results are mixed. Last month Patreon recently laid off about 20% of their staff. Zillow recently had a weak quarter which saw their stock slide 12% today. Tripadvisor tanked yet again as AI overviews and bigger hotel ads in Google SERPs choke the traffic flow to third party review sites. AI rewrites make it easier for Google to redirect more of the click flow & attention to their ads while everything else slides down below the fold.

Cloudflare presumes that the rise of AI agents will lead to widespread AI shopping, but I don't see that taking off the way the hype does. If you are an airline you can use AI to dynamically price seats and get a more efficient yield. That makes a lot of sense because they have so many data points to compare day after day. An individual consumer can not rely on AI the same way because they are doing a single transaction. A consumer's experience within, knowledge of a field, and purchase frequency may be high or low. Demand for quality may be high or low. Need for sustainability may be high or low. Price may be exceptionally important or much less critical. Speed of delivery may be critical. Some will want a one off purchase while others want a subscription with regular restocking. And on and on.

There could be entirely different solutions to solve a problem (rather than buying new shorts I could sew my short pockets, use iron-on sewing tape, or apply liquid stitch fabric glue). Similarly, one could put a beacon tag on their keys or put the key on a larger set which is harder to fall through a small pocket hole. Maybe a locksmith isn't needed and one can just check in the local post office or police station to see if keys were turned in.

Every marketplace has its own quirks and optimizations: "the bottom 80% of men (in terms of attractiveness) are competing for the bottom 22% of women and the top 78% of women are competing for the top 20% of men. The Gini coefficient for the Tinder economy based on “like” percentages was calculated to be 0.58. This means that the Tinder economy has more inequality than 95.1% of all the world’s national economies."

Demographic segments of a market may prefer what others have, while other segments may prefer not to be near certain others. Some recurring problems are tied directly to cultures or genetics. Some judges can't be fair in certain areas.

Each individual has their own ideals they are optimizing for. Chad Ochocinco likes athletes. Those athletes may prefer deeper. Chad Ochocinco's optimization loop sounds tragically narcissistic, and could be off based on child order with common moms vs across partners. More partners also tends to lead to more friction.

If you are buying something for someone else then that layers in other sets of potential idiosyncrasies. My wife took a genetic test showing she is an elite power athlete, which explains why our daughter is so strong. :D

My wife used to play a video game with another player nicknamed Japanman & that sounds a lot like the cute Anpanman, so I have bought my wife a few Anpanman t-shirts. Typically these are a bit hard to find on eBay because they mostly have Anpanman children's clothes. If one uses the Yahoo! Japan auctions there are lower prices and a wider variety of shirts which will fit adult women. Buyers can then ship through Buyee, consolidate purchases at their warehouse, and have the products shipped abroad.

One of my wife's favorite fashion brands is Custo Barcelona. She tends to vibe more with the stuff they made a decade or two ago than their newer designs. Custo Barcelona also created a sister brand with bright colors named Hayes. One of the Hayes designs is of a nurse, which is perfect for my wife's mom since she was a nurse. But sometimes finding those sorts of things is just a random serendipity thing than something you plan out.

Desigual is another fun Spanish clothing brand which sometimes has neat stuff. A more playful but now defunct brand Paramita used to offer lots of playful & colorful designs. They too had a sister brand named Ciudad Campus. There is a whimsical & cartoony Y2K style used by the labels in most of their pieces, including abstract prints.

This is a cute shirt, but $142.50 is a bit steep for it. One could bulk purchase from marketplaces like Sellpy or Micolet & use a service like Ship German as a European mail forwarding analog to Buyee in Japan. If one can buy a few dozen 5 to 10 shirts from the European platforms one could theoretically relist them on Etsy or eBay for like $30 or $40 each and clean up. The Custo Barcelona shirts also tend to go for about twice as much on eBay or Etsy as they do on European marketplaces like Momox Fashion.

The people who think agentic AI is going to replace the process of shopping are likely presuming that people do not like the joy of discovery or the sense of serendipity. Each agent will have some optimizations to it, and those which fan out helping you find options & learn more are going to be more trusted experiences than systems which buy something for you without many layers of feedback along the way.