Now that a few days have passed, which in agentic commerce time is about a whole year worth of updates, I wanted to revisit and spend time going over the updates that I believe to be the most monumental from Google:

Gemini Enterprise for Customer Experience (a LOT going on here)

Most notably:

Alongside the UCP announcement from Sunday, we received a parallel release and announcement of the Gemini Enterprise for Customer Experience, which aims to bring agentic solutions to shopping and customer experience across a large amount of vertical applications for businesses, notably retailers and restaurants as well.

This agentic solution looks to verticalize the entire customer lifecycle journey, from discovery to post-purchase, to returns, support, and I am sure many other interesting abstractions of the agents.

What Google has done is taken agentic capabilities that offer personalization at scale, given those technologies to retailers and businesses in a way that is drag-and-drop to build workflows, noting that "employees of any skill level can launch sophisticated AI with near-zero human engineering," and have done it on top of UCP.

I think that is absolutely incredible.

They are taking the capabilities of tools like Claude Code and Cursor, applying these tools to agentic commerce, and are doing so at an enterprise level. And unfortunately, given that all of these tools across the platform are leveraging UCP, I am sure many different individual agentic commerce shopping, support, sales, etc. agents might no longer be the choice to enterprises given the compatibility of these across their ecosystem, but that is a future conversation that takes place once we understand whether or not enterprises/mid-market are adopting this protocol in mass.

Maybe a few months pass and the announcement and technology were more hype than most believed them to be, but I just can't see that reality. Everyone is trying to figure out how to stay ahead and vigilant, especially with AI-era discoverability, and if Google has developed the protocol to enable this, why not take advantage? More tools in the toolkit.

CX Agent Studio

This IMO is the coolest announcement, and really speaks to the capabilities of drag-and-drop workflows for developing highly personalized conversational agents through a platform that Google created.

And you have to watch the demo that Google uploaded on YouTube. They use AI generated material, not surprising, but the conversation that they showcase with the "Customer Service Agent" of a shoe company speaks to the capabilities of these agents, which are multimodal: speech, video, text.

In the demo with a customer speaking to the agent looking for a better shoe, the agent recommends a shoe based on a video of the runner's gait, its internal knowledge graph, and the customer's purchase history. That is legitimate multimodality.

Example of Google's Business Agent

Example of Google's Business Agent

Not only are you building the agent in the agent studio, but you can evaluate it as well, and then deploy via Google Cloud (not sure about other cloud services). They showcase in the demo an analytics platform for the agents, where you are able to run scenarios, view how the agent performs over multiple iterations of the same requests, and iterate on this historical data. I mean they have really gone the full mile as it relates to shopping agents.

Within the agent personalization are guardrails as well that go over safety preferences/personalization preferences that can prevent dangerous content, harassment, etc., which I imagine for most brands, they would use the guardrails in a strict way, but potentially some brands might be more lenient with the guardrails given the individual brand persona that they promote.

On the Google Agent Studio website, there are links to learn more about creating an agent application, evaluating your agent, and deploying your agent, but unfortunately I was getting a 404 error for each link. I would really love to be able to play within this sandbox and get a better hands on idea of what the drag-and-drop build is like.

They promote it as a solution that does not require an engineering background, and if that is the case, this will allow teams that are outside of engineering's scope to build agents for their organizations, without needing to wait on engineering's priority to meet their demand.

Agent Assist

Agent Assist is also a really cool development of theirs for Customer Support training, enablement, as well as live translation, again trained on a company's internal documentation and knowledge.

They have also included a YouTube demo, which I encourage you to take a look at if it is relevant.

One of the features, AI Coach, is a live conversational feedback agent that can help drive your customer support team to follow conversational flow and support in a way that is aligned with your internal SOPs. The coach is actively viewing the conversation, evaluating the input from the customer, providing suggestions as the output, and even giving suggestions to upsell, which I think is actually incredible.

They're enabling customer support teams to be their sales team, especially after resolving some sort of discrepancy with the customer. This is enabling the frontline IMO.

AI Trainer is essentially a simulated customer interaction that customer support agents can use to train and develop their understanding of the internal knowledge base/SOPs as it relates to interaction with customers. The training provides feedback based on the responses that are given by the agent, which I imagine is incredibly beneficial, especially to larger organizations when they are training their team. Love a cool, simulated environment.

Live translation is exactly what it sounds like. Incredible feature for enabling teams globally to be able to scale up their help to people all over the world. Gone are the days of only being able to work with a customer in your same region.

All together, super cool developments, and again, I am sure there are at least a handful of startups that are going to become obsolete given these developments, but that all comes down to adoption, of course.

Vertex AI Search

This is the part of the stack that quietly turns "AI agents are coming" into "our site actually converts when the query gets weird." And I can only imagine that some of these user queries are lines and lines of text, with different specifications, quirks/requirements, that LLMs need to be good at deciphering.

Vertex AI Search for commerce is Google basically productizing the hard parts of e-commerce search and recommendation into a fully managed, AI-first search layer that's tuned for retail catalogs, not generic document retrieval. Their positioning is pretty direct: "turn search into sales," with personalized and intuitive shopping experiences that are optimized for conversions and revenue.

This is not "semantic search as a feature." It's search + ranking + catalog enrichment + merchandising controls, packaged as something you can actually operate day to day.

Here are some of the capabilities:

If I had to simplify it: Vertex AI Search becomes relevant the moment your shopping experience breaks under modern intent.

Searches like...

"I need a carry-on that fits under most seats, opens flat, has a laptop sleeve, and doesn't scream business traveler" or "protein powder that doesn't upset my stomach, not artificial sweeteners, and ships by Friday."

Google is explicitly pitching this as a system that can serve those channels, not just the site search box.

Takeaways

Product Data

I still think that this is the main takeaway. Agentic commerce is here, it is a defined channel, and companies should evaluate how they can improve their revenue, conversion, etc. in the agentic commerce channel.

If there was any hesitation to work on data hygiene as it relates to SKUs, now is the time to act, REGARDLESS of working within the Google solutions. We understand that prompt queries are so much more specific than the "red fitness running shoe" that someone might be prompting.

Does your product data take this into consideration? How many product attributes can signal agentic discovery to your product? What is the context in which someone might be looking for your products?

Agentic Readiness

Agentic readiness follows the same principles as product/data readiness, but this goes a step further involving UCP, ACP, MCP, etc., taking the time to understand and work with these technologies to benefit your products within agentic discovery in this AI-era.

The easiest first step is making sure that these agents aren't being blocked on your website immediately. Access is the first check on the checklist to ensure that you are at least participating within the agentic commerce channel. If these agents can't access your site, more than likely they aren't accessing your product data, and that is probably intentional.

There are some really cool companies that are operating in the space of KYA, or Know-Your-Agent, which are trying to identify agentic traffic from bad actors such as bots, etc.

Keep Learning

Learning is my last takeaway, but that's more so just a reminder to myself to stay curious and learn as updates come out. There are many, MANY moments where I feel as though I am falling behind, especially in the agentic commerce world, and that's why I will link some resources that I follow to stay up-to-date in the fun, fast-paced world of agentic commerce.

Sources

*posted originally on LinkedIn - link*

Resources