Interesting Agentic Shopping reading this week...
Everyone is waking up to the Agentic + Agentic Shopping wave of innovation and the content is coming fast and furious!
There were some interesting articles and announcements that don’t rise to the level of a single post. Going forward, whenever this happens, we’ll pull them together in an ‘omnibus’ post like this. For this week we’ve put them into three buckets:
Agentic Shopping Minor Announcements
Agentic Shopping Essays and Observations
Interesting Agentic Essays (broader topic, but tangential)
Agentic Shopping Minor Announcements
Opera, a web browser company, announced they are coming out with an Agentic native browser, called Opera Neon, and what caught my eye is this blurb from the press release→
From Prompt to Action
Do is your smart digital operator, capable of navigating the web and completing tasks so you don’t have to.
Neon’s AI agent uses textual representation of websites to understand their content, and interact with it for you, while maintaining your security and privacy. Automate routine web tasks like filling in forms, booking trips, or even shopping.
How’s it work?!
While it’s exciting to see new unpredicted players like this enter the Agentic Shopping Agent arena, since this is just an announcement of something to come, it leaves a lot of questions unanswered. At the same time it gives us some hints at what’s coming up - let’s look at the breadcrumbs
Opera Neon ‘Do’ Agentic Shopping hints
Opera Neon has three agents: Chat, Do and Make. In their microsite for Neon, Do has three cited ‘use cases’→
Plan and book a trip to Lisbon
Find and buy socks
Play the wiki game
For the ‘Find and buy socks’ use case you are taken to a YouTube video that shows it in action→
What the video shows is:
The user prompts Neon Do: “Buy 12 pairs of white socks on walmart. Men's size 10. Have them delivered. I prefer brands like Nike, Adidas or Puma.”
Like ChatGPT Operator, (but vastly different architecture, I get that) you can then watch the browser go to walmart.com
Search for “mens white socks Nike, Adidas, Puma size 10” on walmart.com
On the SERP the first product is Puma 8 pack (remember this is a 12 pack in the prompt), it selects that and goes to PDP
On the PDP we see that this is a 3pm seller, the agent throws it in the cart
Then it’s interesting there’s a persistent ‘thinking’ box and in there, and this is where the 12 pairs from the OG prompt comes in.
The agent knows (via thinking box) it needs more socks, so it goes and grabs a different 8 pack from a different seller. The result of this is not an optimal order that I think a human would execute on:
Pair 1 - $25 (ships June 2)
Pair 2 - $18 (ships June 3-8)
As a human, if I want them fast, I’ll order 2 of pair 1, if I want to save, I’ll order 2 of pair2. The agent here ordered one of each - an unusual choice I’d argue most humans would not make and illustrative of how complicated Agentic shopping is going to get.
Then the video stops before we get to payment. Many browsers have an encrypted store of credit card data so they can do auto form fills for you - I assume that’s what’s happening here and they cut the video because we’d be able to see some digits go by.
Opera Neon Conclusion
In our prediction of Agentic Shopping players post , you may remember this graphic:
I’ll point out that Opera was not on there as a potential player, yet here we are. If we think about ‘browser as agentic shopping engine’, here’s a chart of browser adoption:
At the company level this puts Samsung and Firefox on the map as possible competitors. For Chrome (Google), Safari (Apple), Edge (Microsoft) we now have three more ‘vectors’ they can follow to enter the Agentic Shopping competitive arena.
Also, Perplexity has reported they are working on an Agentic browser called Comet.
Agentic Shopping Essays and Observations
My good friend, Jason Goldberg, has a great white paper that talks about how retailers and brands can prepare for the Impending Disruption of Commerce Search. You can download it here. Highly recommended!
Rex Woodbury, a VC at an early stage firm called daybreak, has a great essay in his substack called Digital Native about
The Information ($ubscription reqd) has two articles about Agentic Shopping. In the first article, “Ai Shopping Race Means a New Talent War is Brewing’. In it they highlight the talent flooding into the Big internet companies to work on Agentic Shopping.
https://www.theinformation.com/articles/leaders-ai-shopping-revolution
The next article is: “How AI and E-Commerce Can Play Nice”. In this piece, they
https://www.theinformation.com/articles/pro-weekly-ai-e-commerce-can-play-nice
Interesting Agentic Essays
I’m clearly a big believer in our broader agentic future and there were two essays by that came out this week in with an ‘anti-agentic’ position. While I disagree with these, it’s always helpful to see both sides of an argument, so I share them in the spirit of saying ’this is what the agentic detractors are arguing’, let’s discuss why they are wrong and how we would refute this point by point.
The first piece is quite negative - the gist of the argument here is if you have LLMs that are probabilistic and non-deterministic, thus they are ‘right’ 90% of the time, and if you chain them together 5x you ge 60% reliability. It’s an interesting argument, but I disagree strongly (one day when I have time i’ll write a counter essay). That being said he lays out his argument very well→
The second is from a substack called ‘The Leverage’ by Evan Armstrong - he’s got an interesting framework of where agents will work and where they won’t and why. I disagree with this one too ;-)
Finally, the company LangChain is one of the earliest agentic framework companies hosted a conference on agentic AI called Interrupt and had a great talk from operator and VC Andrew Ng you can find here→
Enjoy!