Breaking AI Agentic News: Microsoft Copilot Update + Google vs. Pinterest? 🤷‍♂️

CoPilot AI Shopping Product Cards are live and Google to launch a Pinterest killer at Google I/O next week?!

Google AI Agent coming next week at Google I/O?

Next week, May 20-21, Google will host their annual developer conference called Google I/O. The keynote is May 20th at 10am PT / 1pm ET, and if you want to nerd out with me, the livestream will be on YouTube here.

The Information ($$) was the first to break the news in an article by Erin Woo, that ahead of I/O, Google has been previewing an AI coding agent and more relevant to Regailgentic readers, “Pinterest-like feature” that “shows images designed to give people ideas on fashion and interior design'.” They then speculate that

“Google’s upcoming Pinterest competitor could be a play to gain more of a foothold in commercial queries, which are most important for ads revenue. Last month, Sissie Hsiao, until recently the head of Google’s Gemini chatbot, told the same court hearing that ChatGPT had so far only cannibalized Google searches in homework and math, which didn’t generate ad revenue. But Google’s ad chief, Vidhya Srinivasan, believed cannibalization of commercial queries was “inevitable,” Hsiao testified.

I’m going to go out on a limb here and say this makes zero sense to me and has to be wrong. When the rest of their big competitors are paddling as hard as they can to Agentic shopping, I find it very very low probability that Google would be swimming the other way. They’ve had 15 years to compete with Pinterest, why now? Perhaps the early reviewer didn’t realize they were looking at an agentic shopping front end?

If I’m wrong, this Pinterest clone feature at LEAST has to have an AI recommendation hook, but I’m betting that there’s much more to the shopping feature that was previewed including an agentic purchasing flow.

We’ll be covering the event and send out any breaking news.

Microsoft Copilot Product Cards are LIVE!

Last week in our foundational AI Shopping Agent status/backgrounder post we mentioned that on on April 18th, Microsoft announced a Merchant Center for Copilot, which was a harbinger that eventually they would do more in e-commerce.

Today, thanks to subscribers and long-time friend and all around awesome retail/ecommerce analyst, Michelle Grant (Salesforce) we can announce that this is live. Thus we’re moving Copilot fromthe ‘Announced’ to ‘Launched’ …well actually, Launched-ish.

Launched-ish?!

A couple of subscribers mentioned that it’s a bit apples and oranges to compare, say, Perplexity + Amazon’s agents that actually do complete the transaction to ChatGPT Shopping which is clearly only on step 1 of a multi-step rollout - the Research/Find phase - they haven’t implemented ‘Buy’ yet.

Copilot is in the same situation - introducing these product cards isn’t a complete. AI shopping agent, but a big foundational step in the right direction.

Microsoft Copilot Product Cards in the Wild

To get the cards to pop up, so far we’ve had the most success in the shoe category. Here I search for a “Men’s Size 11 Moab 3” and get an initial carousel with 5 ‘styles’ of the Moab 3- 3 low, a mid and a high. It has an error there as it’s duplicated one of the styles and not ‘canonicalized' (or correctly matched/folded together) the 2 of the lower styles.

Here’s page 1 of the carousel

..and here’s page two showing the duplication:

We learned at ChannelAdvisor that people from the search world have a hard time wrapping their heads around the complexities we live with every day in the ecommerce world. This is a classic example. But wait, there’s more!

When you click on a product card, you are now presented with the offers for that SKU from merchants:

As a frequent buyer of this shoe, I know Merrell is pretty aggressive in the pricing here, thus the clustering at $120, except for one notable exception, Amazon. Amazon has some of the most complex PDPs out there and the Amazon variation matrix gives search companies complete fits because instead of it being a two dimensional matrix (size x color) it has a third dimention which is (size x color x seller). Sure enough, Copilot has picked the wrong size, and the wrong color (grey vs. brown) and on the landing page it sends me to the price isn’t the advertised $89, it’s $79.

In fact, Amazon doesn’t have the shoe in the carousel at all in this variation matrix - it’s sold out. Another thing search companies don’t ‘get’ is the need to update attributes such as, oh, availability/in-stock and price. My educated guess is a web crawl > 24 hours ago found the shoe and now it’s out of stock and the landing page flips to something available which happens to be this size 10.5 grey shoe.

Variations a-plenty

Another shoe search, this time for the generic ‘womens running shoes’ caused the product card for this ASICS Gel-Excite 10 to pull up a variation matrix on the card which is interesting because you can clearly see the Merrell Moab 3 did not have that. Here it’s pre-selected size 10 and the ‘White/Light Ube’

Note: The amazon landing page is off again - on both of them. If I move to Size=9 and color=Midnight/Coral Reef, it updates the offers to include 3 asics (the bottom is an ad I believe?) and 2 amazon.

The topic ASICS has a correct landing page (right size/color and price!)

But alas, the Amazon listings were wrong color/sizes and the $85 ASICS was the same listing as above, but it was in twice, once for the OG $85 and then second for the strike thru sale price of $69.95. My guess is there’s a combination of feeds and crawling colliding all in here that’s causing a bit of an ‘out of sync’ experience.

Conclusion

It’s early days and it’s exciting to see Microsoft getting closer to an Agent. They have a Copilot labs agent for buying travel and making reservations (Copilot Actions -pictured below). These product cards feel like the first of a couple steps they’ll need before they go fully agentic.

It was also hopefully helpful for you to see that on the surface with the nice cards and other UX elements, these shopping experiences look very baked and mature, but once you start putting them through the paces, pushing the variation complexity, real-time pricing/avail and other parts of online retail that are more complex, they aren’t quite ready for prime-time. But the prize here is huge and they’ll get there.