Battle of the Agentic Browsers Update: Microsoft Edge enters the Ring with Copilot Agentic Assist - Part 2/2
Part 2 continues our Deep Dive into Agentic browsers
Welcome to Part 2 of a 2 part series where we’re doing a deep dive into the new wave of Agentic browsers. If you missed part 1, you’ll want to start here.
Agentic Browser Smackdown! Microsoft Edge vs. Dia vs. Comet
Now let’s throw the new challenger on the block, Microsoft Edge in with Dia and Comet to see what we see. The topic of this substack is the intersection of Retail and Agentic, so of course we’ll focus on a real-world retail experience.
Use Case: Elsie loves Veggiedents
It all started when I brought our office dog, Elsie, a dog dental chew from home called Veggiedents (according to my Vet they are way better than Greenies and more important to dental health and thus, of course, more expensive!) Now Elsie’s day at the office doesn’t start right if she doesn’t begin the day with a delicious veggiedent. This has caused my Veggiedent consumption to skyrocket and I need to reorder them somewhat desperately. I’m brand loyal here, but want a low price.
An Ecommerce Challenge Worthy of an Agentic Browser
At my previous company, ChannelAdvisor we had a wide range of great, smart, big companies from outside of ecommerce try to tackle ecommerce to find themselves quickly underwater and ultimately retreating. I’ve learned over the last ~30yrs Ecommerce is both incredibly simple on the front-end and to provide that front-end experience, the back end is insidiously deceptively complex.
What makes ecommerce complex? How long do you have? 🤪 Usually the first pitfalls people fall into is variations, real-time inventory, dynamic prices, shipping/tax calcs, kids, bundles, etc. Keep these in mind.
For this context, the setup is I need to reorder some veggiedents→
Between us, I know exactly what I want - I want the ‘small dog’ sku - this is a simple variant. I also specifically want this, product, I don’t want a substitute. Here’s the product from april in my amazon order history→
🤫 Shhh, don’t tell the agent! Part of the experiment here is to make the agent find this as the first in a chain of tasks it will need to do successfully accomplish this task.
The prompt is: (while on the amazon home page logged in with my account within the Agentic browser being tested:
Test Prompt:
“I need to reorder veggiedents -check my amazon order history for that and check the latest price on amazon and then other sites and find me the best deal.”
To pull this off the agent has to:
Go into my Amazon order history
Find the veggiedents
Understand the dog size and weight in ozs component of this SKU
Look for this specific SKU across various sites
Find the best price
Another complexity with this task is this manufacturer seems to be on-again / off-again with various retailers. I’ve gotten them from Amazon, they are off now and they are not on Petsmart or Petco. The prices change a lot and they are hard to find - fun!
Spoiler Alert
The browsers don’t now this yet, but the lowest price is Chewy at $27.59.
Microsoft Edge is Up First
Reminder -you can download the exact browser we use in this competition here, it was released this Monday.
I’m keeping these images pretty large so you can see what’s going on.
Microsoft Edge Analysis
Maybe this is because it’s new, but initially it told me to change some privacy settings to be able to access the order page, and I did that and it consistently told me it didn’t have access. 🤷♂️ To it’s credit, the browser did see on my Amazon homepage Veggiedents (I’d been searching a lot for them via agents thus Amazon surfacing it in my Buy Again’ section). Unfortunately you can see it has erroneously pulled out the Medium size SKU and not the small. (I looked and it’s clearly small).
It then does a decent job of the comparison (of the wrong SKU). You would expect any agent to check amazon, walmart, maybe target, but definitely chewy, petsmart, petco and then some more niche sites if it wanted to be comprehensive.
It did find a half-size and flag it as such - I won’t criticize it for this because it catches and shares that it is half size.
I would have liked it to catch that the min cart for free shipping at Chewy is $35 and this is clearly under, so that should be a signal for it to go calculate shipping.
Where it really kind of goes off the rails (hallucination in LLM-land) is this last little tidbit.
Veggiedents do have a Flex (includes joint health supplement) and Zen (doggie downer version), but the Rufus must have come from Amazon?
How does Dia Do?
Dia is the most mature of our competitors, let’s see how it does:
Dia Agentic Browser Analysis
Dia also gets the size wrong off the bat in the Amazon bullet. On Walmart it’s not very specific at all. On Heartland Vet Supply it gets points for surfacing a retail option I never considered, but when I went to the site to double-verify, Dia didn’t understand that the variation selector changes the price and the Smalls are $27.59 and the Mediums are $31.79
Dia did not seem to check the obvious choices of Chewy, Petsmart, Petco.
Did I mention that ecommerce is tricky?
Comet Goes Last
Finally here’s Perplexity’s Comet Agentic browser tackling this tough challenge:
Perplexity Comet Analysis
First, Comet 100% nails the original SKU on Amazon and as bonus points clicks into the order and figures out what I paid for it back in April. It then looks for the new price and finally (right column of the table) recognizes that’s a pretty big price increase.
Next it does a good job checking Chewy, but doesn’t find the SKU. It’s in a relatively unusual variant matrix here→
What’s unusual about this matrix is if you click small you are then presented with two options for 30 count - ‘30 count bag and ‘30 count’. ‘30 count bag’ changes the size to XS, like a reset of the variation. This is product data ‘jankiness’ (that’s a technical term) in the Chewy product catalog, they didn’t canonicalize the variations correctly, but the agent, like a human, should adapt to that.
Partial credit though as the Chrew ‘human found’ price of $27.59 does land in the range that perplexity gives.
For Walmart, I don’t know what’s up with that - it has flipped to x-small. There is a Small on Walmart from a 3P seller for $35.99, so no harm it didnt’ surface that expensive option, but it loses points for not catching the price swap. What I’m guessing happened is that Comet was specific with the underlying wal-mart search and it returned an X-small (it did text/regex matching, not attribute matching - ugh) and here we are:
Finally, Comet surfaces a new retailer, VetRxDirect, let’s check on that. Comet says they have the small Veggiedents at $24.69, let’s fact check that. Yet again Comet falls for the bad search data and clicks into the search result for a extra-small (notice they use their own way of saying X-small vs. Chewy - yep, ecommerce is hard!)
Sadly the REAL price is $31.79, so VetRXDirect was another bad product-data/search red herring.
Scorecard
Now let’s compare how the three browsers did across 4 criteria ranked from 1 to 10 with 10 the highest, 1 the lowest:
Comet clearly dominated the pack and sadly for this task, Microsoft Edge really didn’t get out of the gate.
Dia did a bit better than Edge, got the order from Amazon, but really flubbed it from there. Perplexity made it the furthest by nailing the first two steps but in the last 2 it got hoodwinked by bad search results and weird variation pages that a human would have figured out.
I will add that this one is a 8/10 complexity - Comet has made it to a ‘40’ on many other tasks for me, but i thought this one was particularly interesting as it highlights some of the real-world ecommerce complexities that those of us deep in the operating layers of ecommerce deal with every day.
Elsie’s Happy Ending (powered by Comet)
At the end of the day, Comet got me 90% of the way there and earned the order which it conveniently placed for me. I ordered 2 packs to get over the free shipping hurdle.
Postscript
I went ahead and ran this query through the other engines and nobody did as well as Comet, BUT I did also run it through the brand new ChatGPT Agent mode in 4o and it crushed it - it found a site, ivet.com that has free shipping over $25, so from a human checking perspective it was the winner here. It took 11 minutes and got the ivet.com cart loaded and ready for checkout where all I had to do was finish the last step.
I will believe agenetic commerce works when it can find best value option for UK rail tickets (which is surprisingly complex in terms of options), or can plan a cross border EU rail trip.
There are a number of dedicated rail travel websites in the UK who generally do OK in rail ticket discovery/purchase (for a fee; fee free from rail line operators but not always best value discovery). And the difference in fare between "dumb" booking and best value is very significant (literally £'00's on some routes).
The most heavily advertised site trainline.com; ($ and also white label as fee-free for some of the line operators without discovery features) is not perfect in best value discovery; and also struggles with European trips. Uber app train booking feature similar to trainline. Others that do better for discovery are trainsplit and trainpal. Seatfrog does last minute fares and upgrades (bit like priceline for flights/hotels)
All websites/apps call back into standard industry systems for ticket issue and reservations, and current/historical service data.
My expectation would be agenetic discovery from first principles consuming currently available open data sets for discovery of timetables and fares; handing off to the most appropriate line operator for fee free fulfilment. Props if they can do multimodal trips using open transit bus data, and coach bus timetables (NX, Flix, Flib, greenline); c.f. traveline.info/moovit
See Jon Worth's blogs about the absolute mess that currently exists for European rail ticketing especially cross border; agentic could help but only so far.