From model upgrade to brand factory: Google is remaking content, insights, and retail workflows in consumer goods

0|Opening: if you make shampoo, a drink, or a snack, why does this update actually matter?

If you have been skimming the technology news at all lately, you have probably seen a few keywords:
Google has launched a new-generation large model, Gemini 3 Pro (G3P);
its image model has been upgraded to Nano Banana Pro (NBP);
and there is an AI platform called Antigravity.

For most mid-to-senior managers in a consumer-goods company, the first reaction is usually: “Another new model, higher scores, a longer context — so what?” What you actually care about every day is whether a new product can win, whether you can close terms with a retailer, whether you are visible on the shelf, and whether the media money is coming back — not whether a model is a few percentage points ahead on a benchmark.

But if you put this update back into Google’s layout over the past two years, what is worth watching is not “the model itself.” It is this: Google is turning AI from a “tool” into a “factory line” that can connect directly to the brand, to content, and to the shelf — from chips and the cloud, to models and agent workflows, and on to the entry points of the phone and the browser.

For a consumer-goods company, the key question is no longer “whether to use AI.” It is:

Over the next three to five years, should part of your consumer data, brand assets, content production, and channel operations grow on a supply chain like this?
Will you treat AI as “a little helper that writes copy,” or will you design it, seriously, as “part of a brand and content factory”?

I|G3P: from a chat model to a “brain for the category and the consumer”

Start with G3P. The practical change for a consumer-goods company is really two things: it can take in the knowledge of a whole business unit at once, and it is set up to “do the work,” not to “chat.”

Until now, most teams have used generative AI in a single-point way: paste a slide here, a few pages of a report there, one interaction at a time. The real meaning of the long-context capability G3P emphasizes is this: you can feed it several years of a business unit’s key materials as a “base knowledge library” — including:

  • brand presentations and the annual brand book
  • consumer insights of various kinds, U&A studies, and tracking research
  • channel and e-commerce digital-shelf data, and conversations from service and social
  • past campaign results and media reports
  • even internal pricing strategy, promotion mechanics, and promotion manuals

In the past, when you asked AI to “help me think of a campaign that can reach Gen Z at the end of this year,” it was a stranger to you.
In future, you can first let it “read through” the history of your category, the brand’s context, and the assets you already have, and only then discuss the next step.
At that point it is no longer an outside creative consultant. It is closer to a “virtual brand manager” that has read all the internal material.

The second shift is that, in Google’s overall product positioning, G3P is emphasized for “reasoning, calling tools, and running a task,” not for keeping you company in a chat. When it is placed on an agent work platform such as Antigravity, what you see is no longer only an answer, but a complete working process:

  • it first reads your historical campaign materials and sales data
  • then it pulls data from a retail-media back end and from e-commerce shelves
  • it does an initial segmentation for you, finds opportunity zones, and draws out several growth paths
  • it produces a draft presentation, charts, and noted assumptions
  • and it leaves artifacts of the whole process, so the team can review, revise, and extend the work

For a brand director or a category manager, G3P is no longer only “write me a summary.” It can be placed in the workflow — you set the question and the guardrail; it first runs through a large volume of work that is repetitive, but still needs a degree of understanding.
You can put your time on deciding which insights are actually worth a bet, which situations should go to an experiment, and which hypotheses should enter next season’s plan.

II|NBP: turning key visuals, packaging, and shelf materials into a content line that can run at scale

Now the image model, NBP. Its most direct link to a consumer-goods company is not “whether it can make a cool piece of art.” It is whether it can turn your brand key visuals, a pile of e-commerce assets, and offline materials into a content line with a rhythm.

First, NBP’s real strength is that it can “see and understand” the visual assets you already have.
You can give it:

  • this year’s main key visual, and the key visuals of the past few years
  • e-commerce PDPs, banners, and livestream preview images from several countries (including Taiwan)
  • offline displays, hang tags, shelf wobblers, and end-cap designs
  • even photographs of competitors’ shelves, posters, and promotional materials

NBP is not only doing OCR. It can understand:

  • which elements in a frame are the lead, and which are only decoration
  • the relationship between the copy and the product, the logo, and the brand’s distinctive assets
  • whether, on the shelf, a person can “tell who you are at a glance”
  • what, in that image, a consumer is most likely to misread or to miss

For design and brand teams, it is closer to a reviewer of visual and information structure:
it can point out that “this version of the e-commerce image has too many claims, and the hierarchy has fallen apart,” or that “this key visual, against the competitor, is actually not easy to catch on the shelf.”

Second, what NBP can help you do is not only “make one image,” but string together the content line a whole campaign needs.
If you give “G3P + NBP,” in one pass:

  • insights and a brief (text from G3P)
  • existing key visuals and the brand’s visual guidelines
  • assets that performed well last year, and the media results
  • the specifications and constraints of several lead channels

what it can help you complete is closer to:

  • a near-usable skeleton of a campaign presentation (including a storyline and chart placeholders)
  • first drafts of key visuals / banners for different channels and formats
  • a suggested priority of messages, and notes on “too much text here, too weak there”
  • suggested adjustments of language and visuals for different markets in Taiwan, Southeast Asia, Europe, and the US
  • a consistency check between offline materials (end caps, price tags) and the e-commerce page

You will still need designers and local teams to tune and to approve the work. But a process that used to take several rounds, and several people passing files among themselves, can be substantially compressed.
NBP’s position is not to replace creative or design. It is to give you a “large production line that can stably produce work at 80,” and to leave the places that truly need 95 for the team to concentrate its fire.

III|From the model to the shelf: Google wants to turn AI into a “brand factory” supply chain

When we look at G3P (text / reasoning), NBP (visual / image), and Antigravity (the agent platform) together, it is not hard to see: what Google wants to build is an AI supply chain that can run from data and knowledge all the way to content, and on to execution.

For a consumer-goods company, if that supply chain is connected, it would look something like this:

  • Data and insights: internal consumer research, e-commerce data, retail-media reports, and service records are first organized into a knowledge base G3P can understand.
  • Strategy and planning: G3P first forms hypotheses, breaks out paths, and runs scenario simulations, and gives the category and brand teams a first draft that has both data and context.
  • Content and materials: NBP takes over, reads existing key visuals, packaging artwork, shelf photographs, and e-commerce assets, and produces a first visual treatment for each channel and situation.
  • Execution and optimization: through an agent platform like Antigravity, “delivery, A/B tests, shelf monitoring, report production, and after-action review” are strung into a workflow that can be repeated.

And at the base, TPUs and Google Cloud are there to ensure that all of this can run inside an existing security and compliance frame, and does not become a pile of “AI experiments scattered across different tools, with no one governing them.”

Once a supply chain like this matures, the question a consumer-goods company faces is no longer only “whether to use G3P.” It is:

Are we willing to let part of our consumer insights, brand assets, and content production, over the next few years, grow inside Google’s AI factory?

IV|Three practical angles for a consumer-goods company: brand, channel, organization

In practical terms, what G3P + NBP suggests for a consumer-goods company can be broken into three layers.

4.1 Brand and innovation: from “one big idea a season” to “a brand brain that keeps running”

On the brand and innovation side, G3P’s most direct value is this: it turns knowledge that used to be scattered across reports, slides, and people’s heads into a brand brain that can be talked to, and that can reason.
That changes several things:

  • onboarding a new person is no longer only “read the slides”; through a conversation with G3P, they can quickly understand the brand’s past campaigns and the traps it has already hit;
  • category strategy is no longer rewritten only in an annual review; as the market moves, it can be adjusted continuously through a conversation with the “brain”;
  • a creative brief no longer starts from zero each time; it is built on cases that have worked, on existing assets, and on consumer context, and gives agencies and internal teams a more solid starting point.

NBP then helps you land what that “brain” is thinking as a visual language that can run across different markets and different shelves.
The real point is not “whether AI can make a good-looking key visual.” It is:

whether a brand can, across dozens of touchpoints, stay clear, stay recognizable, and keep updating with some efficiency.

4.2 Channel and e-commerce: from “one campaign at a time” to “a repeatable content and operations line”

On the channel and e-commerce side, the G3P + NBP combination has a chance to move you from a campaign-by-campaign sprint toward “a content and operations line with a rhythm”:

  • after G3P has read the retail-media and e-commerce reports, it can first “list the opportunity zones” for you: which keywords are worth a bet, which categories have the most elasticity, which combinations work especially well in a given channel;
  • after NBP has read the existing e-commerce asset library and photographs of competitors’ shelves, it can point out that “the information on these key pages is too messy,” or that “these images are beautiful, but they do not help conversion,” and generate a first draft of an improved version;
  • through an agent workflow, “asset generation → listing → tracking → taking down / improving” becomes a fixed rhythm, instead of a human-wave effort every time.

This matters especially for consumer-goods companies in Taiwan and in a regional business:
when the volume of assets explodes, and channel requirements become more and more fragmented, you can no longer respond by adding people in a straight line.
What you need is a system that can “keep producing content that is good enough, and put people on the last 20% of the decision.”

4.3 Organization and governance: from “trying a few AI tools” to “choosing an AI supply chain”

Finally, the organizational layer.
When Google, Microsoft / OpenAI, and other cloud providers are all putting forward their own “AI factory blueprint,” a consumer-goods company actually faces a choice that looks technical, and is in fact deeply strategic:

  • which data can go to the cloud, and which must stay on premise?
  • which processes can be left for an agent to run all the way through, and which must have a person reviewing and a person deciding?
  • which assets can be entrusted to a cloud model to learn from, and which should be handled by a smaller in-house model and internal systems?
  • do you bind deeply to one supplier, or choose a multi-vendor, multi-cloud strategy?

On the surface, these questions are about “technical architecture.” In essence, they are deciding where your organizational capability will grow over the next few years.
G3P and NBP only make the question more urgent:
when AI is no longer only a demo, but can actually connect to brand, content, and commerce processes, you can no longer treat it as an “experimental tool.”

V|Three questions for a senior consumer-goods executive

Finally, no more technology — only three questions, for you who are taking a brand and a category into the next stage:

1|Do you have a “brain for the brand and the consumer”?
Not Google’s, and not necessarily anyone else’s. The question is:
have you already begun to try to turn the reports, slides, campaigns, and data of the past ten years into a system that can be asked a question, rather than leaving them to lie on a file server?
A model of the G3P type is only a reminder: this is now technically feasible. The difference is whether you will start to organize the material.

2|Do your content and your shelves have a “content line that AI takes part in”?
Facing the more and more complex asset needs of e-commerce and retail media, will you keep adding headcount, and tough it out channel by channel?
Or will you start to design a line in which “some assets are first run to 80 by AI, and the brand / design team then supplies the last 20”?
A capability of the NBP type gives you a chance to look at key visuals, packaging, e-commerce images, and shelf materials with one way of thinking, and to update them at one rhythm.

3|Are you now choosing a model, or choosing the “brand-factory supply chain” for the next three years?
You can use G3P today, 5.1 tomorrow, and another vendor the day after. What is truly hard for an organization is:
where are your data and your knowledge being hosted?
which workflow is quietly shaping the way your team works?
whose governance frame will become your default standard for the next few years?

If 2025 is the year the “AI factory officially started work, and began to extend toward the brand and the shelf,” then this G3P and NBP update, for a consumer-goods company, is not a piece of technology news to be read and forgotten. It is a very direct reminder:

AI is no longer only writing a few lines of copy or making a few images. It is growing into a “brand and content factory” you can choose to enter, or not.
The real question is: do you want to, and do you dare to, let part of the brand and category capability of the future grow inside this factory?

The author

David Yang (楊閎), founder of Future Insights Lab

David Yang (楊閎) is the founder of Future Insights Lab. He writes AI Future Reflections (《AI 未來反思錄》), with a focus on Agentic Commerce, AIGC, data methods, and organizational renewal.

Learn more → About the author

More deep-dive pieces

The judgment, restated

My judgment is that a model upgrade is only the raw material. What a consumer-goods company actually has to build is a “brand factory” — content, insights, and retail workflows as one system.

Three things this week

  • Pick one content line, and define the human–AI split for input / checking / publishing.
  • Rewrite an insight need as a question card that can be dispatched (who asks, with what data, and when it is updated).
  • On the retail floor, pick one closed loop: insight → asset → channel → a metric you play back.

Further reading: a close analysis of McKinsey’s State of AI | Google’s multi-agent white paper | a close read of State of AI 2025. Subscribe to Future Insights Lab, or reply and tell us where your brand factory is stuck.

訂閱 Future Insights Lab

不定期寄送 AI × Business 短篇筆記,協助你掌握實務重點。隨時可取消訂閱。