Big Tech might be leading the GenAI race, but specialist companies have an important advantage

Thanks to the boom of LLMs and generative AI, Big Tech is currently in a race to develop the next frontier in AI with intelligent agents that go beyond simply automating tasks, and instead act autonomously on behalf of the user. 

And they're pulling out all the stops to take the lead. News pages are full of everyday use cases coming from these leading AI players, with OpenAI’s GPT4o, Apple’s forthcoming upgrades to Siri, and the hotly anticipated Project Astra from Google on the horizon in 2025, to name a few.

But despite the proliferation of these exciting use cases, I don’t believe the race is yet won.

Is Big Tech’s GenAI victory inevitable? I don’t think so.

Google, Microsoft, Meta and other big names may have the upper hand in the generalist AI market and the broad application of AI thanks to their computational power and access to massive data sets, but I believe niche players have the opportunity to carve out significant roles in domain-specific intelligent agents. Ones in which specialised knowledge and data can lead to substantial competitive advantages in particular fields that generalist systems cannot.

Why? Because smaller, more agile companies that have built up a repository of domain-specific data about their users are better placed to offer more focused, more tailored solutions for their audiences; solutions that are highly relevant to the specific challenges and requirements of their chosen domain.

Yes, it’s true that the likes of Open AI’s ChatGPT and Google will most likely shine across a broad range of generalist tasks (potentially getting much closer to the next frontier of Artificial General Intelligence where AI will have reasoning like abilities that outperform those of a human). But ultimately, I expect they’ll struggle to match the specialised expertise required in fields that are heavily dependent on data and information not available in the public domain: finance, healthcare, even the likes of Strava in the world of fitness, that have private user data already built up in their repositories.

Perhaps this could give rise to a “data marketplace” – where this proprietary and domain specific data is sold to the GenAI provider of the highest bidder. But that’s another topic for another day!

Or more likely, with the combination of deep user knowledge, agility and customisability, these providers can build expert agents in a specific domain, and deliver bespoke capability that is significantly better and more specific than their generalist counterparts. 

Because those already on the journey of building their own knowledge base (for RAG – retrieval-augmented generation) will have quickly realised: if you put garbage in, you will get garbage out.

As Google has already proven when its AI Overview advised users that eating rocks can be healthy, and to glue cheese to pizza, there’s a real challenge in aggregating and interpreting the vast unstructured data sets needed for a comprehensive AI-driven solution in fields like healthcare, finance, legal, etc. - if only to avoid prescribing rocks.  In corporate finance, nuanced insights derived from extensive financial data are crucial for informed decision-making beyond what generalised AI can offer.

And let’s not underestimate the context of the user.

Generative AI will make all of us do things faster in our personal and business lives. If you are a Salesperson who wants to reply to an email in Salesforce, you are likely to ask Einstein to draft a recommended response. This is much easier than copying the email and all the contents into ChatGPT to do basically the same thing. Here, (in theory at least) Einstein has all the data (history of email chain, history of contact with this person, etc.) to be able to do a much better job than if you copy and paste into ChatGPT. Plus, it’s much quicker directly in Salesforce – just one click.

A glance into the future 

Looking ahead, the landscape of GenAI and intelligent agents will evolve dramatically.

Right now, we are still in the early test-and-learn phases, and the technology still has the Gartner “trough of disillusionment” to look forward to. With the rise in teething problems, model hallucinations of some AI tools we’ve seen reports on, and the challenging the reality of implementation, I think many would agree that we are probably entering this phase right now, as our initial excitement inevitably fizzles away.

In the meantime, as we navigate these complexities, the potential for AI to enhance productivity and decision-making across industries remains vast, promising a future where human-machine collaboration will reach new heights of efficiency and innovation.

And with it, I believe smaller players still have a big set of shoes to fill, and the need to build proprietary data sets has never been greater.

In the future, I’m certain that generalist and domain-specific AI intelligent agents will co-exist, each serving distinct roles. But there’s an exciting world of opportunities ahead… so who’s ready to come out on top?