CTO Perspectives: An AI Reality Check

AI… It's a tide of hype, promises, and some genuine breakthroughs. 

In both the technology and consulting markets, everything is about AI right now, and all movements should be examined in this light. 

If you're in sales or management, it can be tough to tell reality from noise. My goal here is to give you a straightforward look at the market. We'll cover where AI truly stands today. This is part one of our series to get you up to speed.

From "Wow!" to "How?"

According to industry analysts at Gartner, AI has, by and large, reached the "Peak of Inflated Expectations." (This wave of AI, that is.) Now, we're entering the next phase. The conversation has shifted from "Wow, AI can do anything!" to "Okay, how do we actually make money with it?"

The Gartner Hype Cycle for AI, 2025. Source: © Gartner, Inc. and/or its affiliates. All rights reserved.

Compared to last year, pragmatism is taking over. 

The most notable trends

• Generative AI (GenAI) has moved past the peak of inflated expectations toward more realistic use cases with tangible ROI.
• Interest in AI agents and AI-ready data is growing. 
• AI-driven solutions move from labs to standard business processes, and it doesn’t work in “cowboy style” in production: AI Engineering and model operationalization (ModelOps) are key to creating and scaling sustainable AI solutions. 
• AI-assisted coding has strongly impacted the developer market and developer productivity. This has been, perhaps, the most visible and ubiquitous use case so far.
• The focus is shifting to practical applications and scalability, while organizations must manage ethical and regulatory challenges.
• Today, "document-chat" type applications are self-evident; the novelties are moving towards agents (or agentic systems) that perform tasks on their own. This also means that RAG applications can now be reliably developed and implemented in organizations; those who want to be on the leading edge experiment with agentic systems and more robust AI ontologies.

In my opinion, the hype will gradually turn into pragmatism, and "AI" will become a part of normal life. I definitely don't expect a "one AI to rule them all" in the near future. Artificial general intelligence is less of a key issue for the economy right now. The important milestone is a sufficiently stable, strong enough, "purpose-driven", practically applicable AI in the workplace. In many areas, we are already there; in other areas, we are quite close; in certain complex topics, we are quite far. But AI, as in LLMs, are already "good enough" for many business applications. 

Most major innovations were quite immature and rough at their first applications. They got refined, sometimes even significantly altered, along the way, before reaching their mature status (which, itself, is rarely the ultimate one). Think about firearms, combustion engines, electricity, airplanes, etc.  Whoever applied them whenever they became fit for a purpose got ahead a lot compared to those waiting for the innovation to get more sophisticated.

Generative AI (and AI in general) is no exception.  It’s not “there” yet, we have not had AGI, and much less general superhuman intelligence. It’s something we can think about and even make bets as to when it will “appear”.  But a more practical question is “How can we make use of what we already have today?”.

As more and more mature solutions are engineered around LLMs, monetization will strengthen further. 

Also, I strongly believe that the general public focus will shift from core technologies to their applications. The question will not be "which model" to use, but "what for" and "how". If the use case is well thought out and the software works, there will be much less pressure to replace the underlying models with every new release. 

AI is to become more deeply and intensively integrated into a growing part of business solutions (both products and services), and into corporate operations. There will be AI in everything. Initially often "blindly," and many times for the sake of riding the hype, but after a while, gradually, it will be mostly useful. At the dawn of microelectronics, radios were marketed with the number of transistors they had; and who cares about it anymore? Even portable radios, themselves, are a rare sight these days, but it is still common to get to music and news via radio waves. That’s how I think today’s AI is to evolve.

Having said all this, let me also gamble with my chips: I predict that the pace of foundational research in AI in its current form is more likely to slow down than to accelerate. There will be new results, perhaps even significant ones along current directions. There will be more fine-tuned, smaller models for more niche applications. But for a quantum leap towards a more complete AI, I think we will need new paradigms: LLMs are to be an important part but I am quite certain that language and the way it is modelled via LLMs in and of itself is not sufficient to represent a complete view of the world’s realities. This new paradigm may already be emerging–but I cannot yet put my finger on it. Thus I do not expect a paradigm-shattering breakthrough in the next 2-3 years (e.g., a widespread, "good for everything" AGI). Note, though, that LLMs have taken us quite far, farther than most (me included) had expected; and the field is very volatile at the moment. So I may well be wrong. 

We will see.

In the next parts of this series, I'll take a look at how the world of data platforms, data-centric application development and analytics engineering is changing. So stay tuned!