Welcome back to Neural Notes, a weekly column where I look at how AI is influencing Australia. In this edition: Meta launches a new AI model that is taking aim at shopping.
Related Article Block Placeholder
Article ID: 334415
After abandoning a costly metaverse push, Meta Platforms is scrambling to regain ground in the AI race. Overnight, it announced Muse Spark, the first model from its somewhat newly-formed superintelligence team, alongside a clearer strategy for monetising AI at scale.
While the topline news is that Meta has built a multimodal model designed for everyday use across its apps, perhaps the more interesting part is how strongly it seems to be tied to commerce and discovery inside its ecosystem.
According to its announcement blog post, Meta is positioning Muse Spark as “purpose-built for Meta’s products”, with the model already powering its standalone Meta AI app and website. It’s also set to roll out across Instagram, Facebook, WhatsApp and Messenger.
“This initial model is small and fast by design… a powerful foundation,” the company said.
Smarter business news. Straight to your inbox.
For startup founders, small businesses and leaders. Build sharper instincts and better strategy by learning from Australia’s smartest business minds. Sign up for free.
By continuing, you agree to our Terms & Conditions and Privacy Policy.
Unlike rivals such as OpenAI, Anthropic or Google, Meta isn’t trying to compete purely on model performance or developer adoption. It seems to see its advantage in a baked-in audience. By embedding AI directly into platforms already used by billions, it has the opportunity to shape how people discover and evaluate products without ever leaving the app.
One could argue this means it doesn’t need overall model dominance against other big tech businesses that have pushed into the AI space.
AI becomes the storefront
Meta has been upfront about its commerce plans.
Muse Spark is introducing a “shopping mode” that pulls from content across Instagram, Facebook and Threads to recommend products, styles and brands.
Related Article Block Placeholder
Article ID: 331137
Rather than directing users to external search engines or marketplaces, Meta is keeping that journey inside its ecosystem. Recommendations are shaped by creators, posts and community activity — not just product listings.
Meta says the system will “surface ideas from the creators and communities people already follow”, effectively turning social content into a recommendation engine.
Multimodal AI changes how customers interact with products
Muse Spark is also built around multimodal inputs, allowing users to interact with the model using images as well as text.
In practice, that could mean snapping a photo of a product to compare alternatives, estimating nutritional information from a meal, or visualising how an item might look in a space.
This brings Meta closer to tools like Google Lens, but within social and messaging apps. The implication is that discovery becomes more visual and less reliant on structured product data.
The model’s ability to run multiple “agents” in parallel also mimics the overall market shift towards AI that completes tasks.
On performance, Muse Spark appears competitive but not leading, placing around the middle of leading AI benchmarks in early evaluations. It performs strongly in areas like language and visual understanding, while still trailing in coding and more complex reasoning tasks.
Even Mark Zuckerberg has tempered expectations, suggesting early models will be less about immediate dominance and more about demonstrating progress.
Related Article Block Placeholder
Article ID: 330249
Perhaps the most interesting thing about this announcement for small businesses is that Muse Spark isn’t the model itself, but where Meta is placing AI in the customer journey.
Discovery is moving inside platforms, where recommendations are shaped by content, creators and community activity rather than traditional search. At the same time, commerce is being pulled directly into conversational interfaces, shortening the path from inspiration to purchase.
Similar to other AI platforms, such as ChatGPT, that have pushed into e-commerce, what drives visibility here is important.
It’s no longer just about ranking in search results or running ads, but about whether products are embedded in the kinds of content Meta’s systems can interpret and recommend. Brands that already have a strong presence across Instagram and Facebook are likely to be better positioned.
The move toward multimodal AI adds another layer. As users begin interacting with products through images as well as text, how items appear in photos, videos and user-generated content will carry more weight. Recognition, consistency and context all become part of how products are surfaced.
At the same time, Meta’s shift towards more controlled access suggests businesses will be working within the platform’s rules rather than building directly on top of its models. Understanding how content is ranked and recommended may matter more than accessing the underlying technology itself.
Because if Meta gets this right, the next version of “search” on its platforms could be more like a conversation inside the apps your customers are already using.
While the new features are rolling out first in the US, broader availability (including across Meta’s core apps) is expected to follow.
