Are AI Agents Fashion’s Next Most Powerful Tool?

This week, Google held its annual Cloud Next event, a three-day conference in Las Vegas during which the tech behemoth unveils all sorts of new cloud-related capabilities. Some have speculated that the move by Google is part of a larger trend by the “Magnificent 7” (formerly known as FAANG) to push further into the B2B space, given that Microsoft has been the dominant player here for a long time – even if Google Workspace has made inroads. 

Among the avalanche of generative AI releases and previews unveiled at the conference, there were two highlights that we noticed: Vertex AI Agent Builder and Creative Agent – both form part of Google’s enterprise Vertex AI platform. The latter is an image-and-text-generation bot,  pitched at marketing and in-house creative teams, and intended to compete with the AI portions of suites like Canva. 

Creative Agent helps creative teams to craft and customise content using generative AI, and also offers capabilities for brainstorming, storyboarding, generating social media content, and even podcast scripts and voices. Platforms like Adobe Creative Cloud and Sitecore also provide tools with comparable functionality, but Google has emphasised its unique value proposition by integrating this feature into its Workspace platform, ensuring usability for a non-technical audience. 

The creatives testing new AI offerings have so far been positive, even if they have to be as they were given first access, as in the case of Sora. You have to wonder, why would creatives – designers, illustrators, advertising and marketing personnel etc – happily adopt the tools that could potentially automate a large part of their jobs and leave them irrelevant? The current consensus – although this is far from proven in practice – seems to be that creatives will continue to bring original ideas to the table, and technology will then expedite the execution process.

Tool-assisted workflows are obviously nothing new in the artistic / creative space, where the transition from analogue to digital processes has significantly moved the demarcation lines between human effort and machine assistance. But while having software act on a person’s behalf is familiar in some sectors, it still represents a new frontier in professional and personal AI – with the new, idealistic, vision being that users will interact with AI chatbots, who will then accomplish tasks for them.

If this vision pans out, then Google’s Vertex AI Agent Builder could prove to be very popular, in the fashion industry and beyond, since it promises a turnkey way to build AI interfaces that are then empowered to take limited actions (in this case within Google’s domain). In the company’s own words, this tool “lets developers easily build and deploy enterprise-ready gen AI experiences via a range of tools for different developer needs and levels of expertise – from a no-code console for building AI agents using natural language to open-source frameworks like LlangChain on Vertex AI.”

This evolution from basic chatbots to conversational agentic AI is, in some people’s opinion, the next frontier of computing. On a personal level, we’ve all previously interacted with assistants to set timers, or predict the weather, but we’ve also all likely come away frustrated by those assistants’ limitations – the points at which we need to step in and “just do it ourselves”.

In personal software, moving those points further away from the user could prove to be a genuine shift in how we interact with technology. Something on the same scale as the introduction of smartphones. Asking an AI agent to edit a photo for you is an easy example; asking one to book you a hotel, or organise an entire trip itinerary is a much harder one.

In enterprise software, moving those points will require a much higher bar to be hit. AI agents, if they’re going to act on behalf of creative, commercial, technical, management, and other professionals, will need to be as accurate and trustworthy as possible – two areas that generative AI models notably struggle at the moment.

To address this to some extent, Google’s new toolkit brings together the company’s latest Gemini large language models (LLMs) and incorporates both Retrieval Augmented Generation (RAG) APIs and vector search techniques. These methods are widely adopted in the industry to mitigate “hallucinations” – instances where models generate incorrect responses when accurate answers are out of the scope of the data that the model was trained on, or where inference leads to the famous “confidently wrong” track that AI models sometimes go down.

This could be welcome news for fashion companies who have been looking for an AI solution that is more reliable than what is currently available, to keep AI applications and experiences on-brand.

With AI agents, in theory, fashion brands could integrate model capabilities with external systems, enabling hyper-personalisation and the execution of tasks on behalf of their customers. These agents might also analyse past collections and marketing campaigns to grasp a company’s identity, and use this understanding to generate new ideas consistent with the established style. By training on brand images, descriptions, videos, and documents, AI agents can generate content like images, captions, and other materials that align with the company’s unique style – as well as having the ability to take action on users’ behalfs.

But of course, building the model for a decent AI agent will require a lot of data for training. And under the current rubric this means giving even more information to big tech companies. And some will argue that the true concern when it comes to AI isn’t the technology itself, but rather from the possibility of its monopolisation by tech giants and governmental entities. The centralisation of AI by Microsoft especially is currently a concern for a lot of people who want to see the future of technology benefit everyone.

Against this backdrop, there are a growing number of companies working on decentralising AI – creating infrastructures that back decentralised AI data inference and a network of community-owned small language models (SLMs). SLMs are a targeted approach to AI, aiming to address specific use cases with greater efficiency and lower costs compared to their larger counterparts. Whether the fashion industry would prefer this remains to be seen.