DaVinci Commerce, the company formerly known as Jivox, on October 5, 2026 put a product called DaVinci Brand Agent on the market, describing it as a system that builds a persistent layer of context around every product a brand sells and hands it to AI shopping agents in formats including ACP, UCP, MCP and JSON-LD. The company says deployments with "several large global brands" have produced "over 30% increases in product discoverability across many LLMs" - a figure published without a baseline, a method, a time frame or a list of the models tested.

In Short

A company called DaVinci Commerce built a tool that writes detailed notes about each product a brand sells, such as who it suits and when it is useful, and passes those notes to AI assistants that shop for people. It matters to brands and retailers because AI assistants like Meta's Muse are starting to pick products on a shopper's behalf, and a product the assistant does not understand may never be suggested. The company says the tool made products more than 30% more discoverable, but it has not explained how that was measured, so the number cannot be checked.

What DaVinci Commerce published

The announcement arrived as a blog post titled "Introducing DaVinci Brand Agent: Making Products Discoverable in the Age of Agentic Commerce," dated October 5, 2026, written by Diaz Nesamoney and filed under the site's Thought Leadership category. Nesamoney is founder and chief executive of DaVinci Commerce. The post opens by stating: "This morning we announced DaVinci Commerce Brand Agent, powered by our patent-pending Product Context Memory (PCM)."

That morning announcement was a press release distributed through PR Newswire from San Mateo, California, at 09:00 Eastern time on the same date. The release carries more technical detail than the blog post. It also gives an availability date of October 1, 2026 - four days before the announcement itself. Neither document mentions pricing.

The two documents name the product slightly differently. The headline and closing line of the blog post use "DaVinci Brand Agent," while the opening sentence and the architecture diagram use "DaVinci Commerce Brand Agent." Both carry a trademark symbol. The page metadata on the company's site shows a modification date of October 20, 2026, fifteen days after publication; the post does not say what was changed.

Why the company says product catalogs fall short

Nesamoney's argument begins with a shift in how purchases are made. "We are rapidly moving from AI-assisted shopping to agentic shopping," he wrote. "With the emergence of Personal AI Agents such as Meta Muse, ChatGPT dots and Instinct, consumers will increasingly delegate tasks to AI, from booking hotels and ordering food to researching, selecting and ultimately purchasing products."

The problem, as framed in the post, is that brands and retailers do not know how their products get selected when software makes the recommendation. Nesamoney then distances the product from the category of tools that has dominated marketing conversation about AI search over the past year. "I don't believe GEO alone solves this problem," he wrote.

His case rests on the structure of product data. "Traditional product catalogs were built for search. They describe what a product is: its attributes, specifications, ingredients and features," according to the post. "Consumer Agents need to understand why, when and for whom a product is relevant."

The example he chose is a sneaker. A shopper might need one for comfort, a particular activity, a foot condition, travel, weather or style. "The same product could potentially satisfy thousands of different consumer needs. A single optimized product description can't capture all of that," the post states.

It is a reasonable description of a real gap. Product feeds, the files merchants send to Google, Meta and others, are built around fixed fields: title, price, availability, brand, category, image. A language model answering a question like the one shown in DaVinci's diagram - "What would be a good foundation for me? I have acne-prone sensitive skin." - needs information that rarely sits in those fields. Whether any agent actually reads a brand-supplied context layer to answer that question is a separate matter, and one the announcement does not address with evidence.

How Product Context Memory works

Product Context Memory is the core of the release. According to the blog post, it is "a dynamic, persistent product context layer purpose-built for the agentic commerce era," which "continuously builds and maintains rich context around every SKU, connecting authoritative product information with consumer needs, benefits, use cases, occasions, preferences and supporting evidence."

The architecture diagram in the post labels PCM a "dynamic knowledge graph of product data." According to the press release, PCM organises its per-SKU knowledge into a knowledge graph and uses semantic indexing for retrieval. The company has filed a patent application; neither document gives a filing number or jurisdiction.

Three components

The press release places PCM inside a larger product it calls the Agentic Discoverability Engine, which has three parts:

  • A Product Enrichment Engine, which uses research agents to identify needs, benefits, use cases, occasions, preferences and evidence for each product.
  • Product Context Memory, which holds and updates that context over time.
  • Product Discovery Intelligence (PDI), which measures discoverability by brand, category and SKU.

Brand Agent sits on top of this stack as the publishing layer. In the diagram, its box lists four functions: "Closed-loop attributed discoverability measurement," "Research agents power contextual product enrichment," an agent that "publishes product content via open protocols," and Product Context Memory itself.

Inputs

The right side of the diagram shows where the raw material comes from. Under the heading "Commerce Platforms," it displays the logos of SAP Commerce Cloud, Adobe Commerce Cloud, Salesforce Commerce Cloud, Shopify Plus and Bazaarvoice, along with a generic PIM/PXM icon. A column labelled "Product & Research Agents" lists six input types:

  • Product Catalog - SKUs, pricing, availability
  • Enrichment Inputs - use cases, FAQs
  • Reviews & UGC - ratings, reviews, Q&A
  • Promotions & Loyalty - offers, personalisation
  • Commerce Services - checkout, fulfilment
  • Customer Support - returns, status, Q&A

The diagram carries a caption stating "Trademarks owned by their respective owners." No document describes these as formal integrations or certified connectors. The press release says enriched content can also flow back to retailer product detail pages through existing PIM and commerce infrastructure, which suggests the platforms are read from and written to, though the mechanism is not specified.

Protocols and formats

"DaVinci Brand Agent makes that enriched product knowledge available to AI agents in the machine- and agent-readable forms they increasingly use including ACP, UCP, MCP, JSON-LD, structured product feeds and enriched retailer PDP content," according to the blog post.

The diagram, titled "Personal AI Agents Speak To Brand Agents," assigns each format a role under the heading "Machine-Readable Protocols - Standardized ways for consumer agents to speak to brand agents":

  • ACP - product discovery, cart, checkout
  • UCP - catalog, pricing, loyalty, personalisation
  • JSON-LD - product details, offers, reviews, rich context
  • MCP - conversational tools, brand knowledge

The selection covers the main standards now in play. ACP, the Agentic Commerce Protocol, was released by OpenAI and Stripe on September 29, 2025, alongside Instant Checkout in ChatGPT. The UCP, or Universal Commerce Protocol, was published by Google on January 11, 2026 with Shopify, Etsy, Wayfair, Target and Walmart, and in April Amazon, Meta, Microsoft, Salesforce and Stripe joined its Tech Council. MCP, the Model Context Protocol, is a general method for connecting language models to external tools and data rather than a commerce standard. JSON-LD is the format most sites already use to embed structured data in product pages for search engines.

That mix matters. Publishing to a protocol is not the same as being read by an agent. Each standard has its own adoption curve, and the agents named in the announcement do not all support the same ones. The announcement does not say which formats are live for which agents, or whether any agent operator has agreed to query DaVinci's endpoints.

Agents named, and what the documents do not say

The diagram's left column, "Consumer Shopping Agents," describes how "Consumers discover, compare, & shop via AI assistants: listings, overviews, apps / plugins & more." It shows the logos of ChatGPT, Gemini, Claude, Meta Muse, Alexa and Sparky, Walmart's shopping assistant. The press release lists the same territory in three groups: large language models (ChatGPT, Gemini and Claude), personal AI agents (Meta Muse, OpenAI dots and Instinct) and retailer shopping agents (Walmart's Sparky and Amazon's Alexa for Shopping).

Neither document says whether these are partnerships, technical integrations or live deployments. The more likely reading is that they are the surfaces DaVinci's output is meant to reach, through the open formats above. Nor do the documents explain "ChatGPT dots," which the press release calls "OpenAI dots." No PPC Land coverage of a product by that name turned up in searches.

Meta's Muse, the most prominent name on the list, is a recent arrival. Meta put Muse inside WhatsApp and a standalone app on September 8, 2026, less than a month before DaVinci's announcement. Within days of its release, Amazon blocked the agent from shopping on Amazon.com, citing among other grievances its failure to identify itself while browsing. Such disputes show that access for personal agents is not settled - and that what a brand publishes may be read on terms set by the agent's operator, not by the brand.

The 30% claim

The only performance figure appears in a single paragraph of the blog post. "And we're already seeing the impact. The technology is being deployed with several large global brands, with over 30% increases in product discoverability across many LLMs in many of our deployments," Nesamoney wrote.

Each qualifier narrows the claim. "Over 30%" is a floor, not a measured average. "Many LLMs" does not say which, or how many. "Many of our deployments" implies that some deployments did not reach that level, without saying how many did. The brands are unnamed. No baseline period, prompt set, sample size or definition of "discoverability" is given.

The press release, issued the same morning, contains no 30% figure and no quantified result of any kind. It does describe how the measurement layer is supposed to work. According to the release, Product Discovery Intelligence tracks retrieval, recommendations, mentions, relative rank, competitive share of voice and adoption of enriched content. It supports baselines, gap analysis and A/B tests comparing raw and enriched catalogs. The release also describes "attributed discoverability," which measures whether AI-generated mentions lead consumers to brand or retailer product pages.

These are the right questions to ask of such a system. An A/B test of a raw catalog against an enriched one would produce exactly the kind of evidence that could support a 30% figure. The company has not published one.

There is a structural problem with any discoverability metric of this kind. Language model answers vary from run to run, change with model updates and depend heavily on how a question is phrased. A rise in mentions measured by the vendor that sells the enrichment, on prompts the vendor selects, is a vendor-supplied figure and has to be read as one.

Named customers

The press release lists Nestlé, Diageo, Giant Eagle and Nordstrom as DaVinci Commerce customers, "and many more." The same four names appeared in a DaVinci Commerce release on March 23, 2026 about an earlier product. Neither document says that any of them uses Brand Agent or Product Context Memory, or that any is among the "several large global brands" behind the 30% claim.

From Jivox to DaVinci Commerce

The footer of the company's site reads "Jivox Corporation DBA DaVinci Commerce 2008-2026," and its login button leads to app.jivox.com. DaVinci Commerce is, in legal terms, Jivox operating under a new name.

Jivox built its business in personalised advertising creative. A Forrester Wave for creative advertising technologies named Jivox a leader alongside Innovid, Clinch and Flashtalking. In July 2021, Jivox updated its Dynamic Canvas Studio, a tool for generating thousands of video ad variations from product feeds and digital asset management systems; Nesamoney, then described as founder and CEO of Jivox, was quoted in that announcement.

That lineage explains the shape of the new product. Dynamic creative relies on taking a product feed and producing many variants of an ad for different audiences and contexts. Product Context Memory applies a similar idea to the feed itself: one SKU, many contexts, each one matched to a different consumer need.

The March 2026 release disclosed a strategic investment by Accenture, without an amount, and a partnership under which Accenture would integrate DaVinci Commerce into its commerce, data and AI offerings. That release launched the Agentic BrandStore, a conversational brand storefront inside LLMs, beginning with a ChatGPT app. Brand Agent is the company's second agentic product of the year. The site's navigation also lists Commerce Media and Commerce Content as product lines.

The blog's author biography says Nesamoney founded three technology companies, and that he was previously co-founder, president and chief operating officer of Informatica, which he "took from a startup to a publicly traded company in 1999." According to the biography, Informatica "was acquired by Salesforce for $8 billion in November, 2025."

Why this matters for marketers

The announcement lands in a market where the plumbing of agentic shopping has been rebuilt several times in twelve months. OpenAI's Instant Checkout went live in ChatGPT with Stripe in September 2025. By March 2026 it had been shut down, after Walmart reported that purchases made inside ChatGPT converted about three times worse than click-outs to its own site. In June, OpenAI moved product feed management into ChatGPT Ads Manager, turning the catalog infrastructure built for checkout into an advertising input.

Protocol adoption has been thin. In May 2026, Originality.ai found only 26 UCP implementations across more than three million sites. Google's response has been to adopt on merchants' behalf: in September, it switched on AI Mode checkout for Shopify stores without asking. The agent operators, in other words, have not waited for brands to publish to their standards.

Brand Agent is also not the first product built on the premise that standard feeds are too thin for AI. In March 2026, Botify put Agentic Feeds into tech preview, enriching product data with reviews, Q&A and other context, with one unnamed retailer feeding more than a million products through it. Measurement vendors have moved in the same direction; in December 2025, Evertune brought its AI brand-visibility reports into impact.com's partnership platform. The pitch from DaVinci is that context, not optimisation of a page, is the missing piece - a sharper version of an argument the GEO sector has been making with less specificity.

Then there is the question of who pays to be chosen. Koddi data from June 2026 framed the AI shortlist as commerce media's next paid placement, with only 20% of consumers comfortable with fully autonomous AI shopping. If agent operators sell those shortlist slots, the value of brand-supplied context to an agent could shift from a ranking input to something closer to ad creative. DaVinci's own product line - Commerce Media sits next to Brand Agent in its navigation - straddles both possibilities.

For agentic AI shopping to rely on brand-built context layers, agents have to trust brand-supplied claims about who a product suits. The announcement's own diagram shows "supporting evidence" flowing in from reviews and research agents, which may be intended as the answer. It does not say how an agent would distinguish a brand's evidence from a brand's marketing.

What the announcement leaves out

Set against what a buyer or analyst would need, the gaps are substantial:

  • Method for the 30% figure. No baseline, prompt set, model list, time frame, number of deployments or definition of discoverability.
  • Which agents consume the output. Logos and names appear, but no document says that any agent operator queries Brand Agent or has agreed to.
  • Protocol status. No statement of which formats are currently live, or which agents read which formats.
  • Customers. Named customers are listed for the company, not for this product.
  • Pricing. Not disclosed.
  • Patent. Described as patent-pending, with no application number or jurisdiction.
  • Governance of claims. No description of how health, safety or suitability claims generated by research agents are reviewed before publication - relevant for categories such as the skincare example the company chose for its own diagram.

The availability date is also odd. The press release says Brand Agent was "available beginning October 1, 2026," four days before the announcement, while the blog post states the company is "excited to bring DaVinci Brand Agent and Product Context Memory to market."

The post closes with the line that sums up the bet. "The next generation of commerce won't just be about getting consumers to find your products. It will be about making sure their AI agents choose them."

Timeline

Summary

Who: DaVinci Commerce, the trading name of Jivox Corporation, and its founder and chief executive Diaz Nesamoney, who wrote the announcement post. The intended users are brands and retailers; the intended readers of the output are AI agents including ChatGPT, Gemini, Claude, Meta Muse, Instinct, Walmart's Sparky and Amazon's Alexa for Shopping, none of which is described as a partner.

What: DaVinci Brand Agent, built on a patent-pending Product Context Memory that builds a knowledge graph of context around each SKU and publishes it through ACP, UCP, MCP, JSON-LD, structured product feeds and retailer product pages. The company claims "over 30% increases in product discoverability across many LLMs" in many deployments, without a published method; the same-day press release contains no figure.

When: Announced on October 5, 2026, with availability dated October 1, 2026 in the press release. The blog post's metadata shows a modification on October 20, 2026.

Where: Announced from San Mateo, California, through a press release and a post on the company's website. The product draws on commerce platforms including SAP Commerce Cloud, Adobe Commerce Cloud, Salesforce Commerce Cloud, Shopify Plus and Bazaarvoice, according to the company's diagram.

Why: DaVinci Commerce argues that product catalogs describe what a product is, while AI agents need to know why, when and for whom it is relevant. With personal shopping agents such as Muse now live, brands face the prospect of software choosing products on a shopper's behalf, and the company is betting that richer, machine-readable context will decide which products are picked - a bet whose central metric it has not yet backed with data.