The risk of letting GenAI define the future of AI

Organisations are being asked to deliver real outcomes from AI, but the hype surrounding generative AI is making it harder to distinguish where real value can be delivered.

Much of this challenge comes down to GenAI and AI being treated as interchangeable. Conflating the two isn’t just a matter of terminology; it can generate significant strategic risk. In practice, that risk plays out in how organisations prioritise and deploy AI. Failing to distinguish between the narrow application field of GenAI and the much broader and powerful AI discipline, risks putting both their transformation ambitions and critical operational capabilities at risk.

AI is a computer engineering discipline that has been around for more than 70 years, spanning a range of techniques, such as machine learning and agent-based systems, that already operate reliably at scale across enterprise environments.

GenAI, by contrast, specifically focuses on generating new content, strategies, methods and designs by learning from large repositories of original source material. It is a powerful capability but only one application field within a much broader discipline.

Recognising this distinction is essential to ensuring GenAI is applied where it adds value, rather than letting the hype eclipse the greater promise of the AI discipline as whole.

Reliability is the dividing line

One of the fundamental differences between the operationalisation of foundational AI techniques and GenAI practice revolves around reliability. This is a critical quality when deploying IT systems at scale.

Reliable AI systems perform with enough accuracy across real-world tasks, behave consistently across repeated attempts, maintain robust performance when conditions change, fail in predictable and identifiable ways and limit the impact when things go wrong.

Foundational AI techniques have demonstrated reliability over thousands of use cases and years of operations. They underpin systems that run at massive scale daily, from aviation and payment to marketing, fostering trust and enabling automation.

GenAI has changed that. The degree of reliability of GenAI models, although respectable, isn’t yet at the level of their foundational AI counterparts. At the same time, large language models (LLMs) suffer from a lack of explainability and transparency, limiting where GenAI can be trusted to autonomously operate at scale and deliver return on investment.

The point isn’t the untrustworthiness of GenAI practices, but how to make it more trusted to the level of foundational AI techniques, so their combined power can be safely applied at scale.

Radically different economic dynamics

The way organisations pay to leverage GenAI models is also radically different from foundational AI models.

Vendors have embedded or added GenAI capabilities into their applications to drive revenue. This often results in higher costs and increased complexity, heightening the risk of unplanned expenses and financial exposure, and making it more difficult to scale AI initiatives efficiently.

Organisations that don’t actively impose structure when purchasing GenAI risk significant financial and operational setbacks. Managing GenAI, therefore, requires not only technical discipline, but clear governance over commercial models and cost.

Agentic AI raises the stakes

GenAI isn’t the only AI practice reshaping the landscape. Agentic AI is another, further reinforcing why these distinctions matter.

Agentic AI and GenAI are separate practices. They can leverage each other to great benefit, but also to great peril: calling it by another name doesn’t make GenAI more reliable.

AI agents are autonomous or semi-autonomous software entities that use AI techniques to perceive, make decisions, take actions and achieve goals in their digital or physical environments.

Agent-based computing is a software paradigm that has been around for decades, and these systems have long relied on foundational AI techniques reliably.

What differentiates agentic AI is its ability to orchestrate decisions using multiple AI techniques. Agents can draw on foundational AI as well as techniques that underpin GenAI models, leveraging parallelism and distributed optimisation to execute complex tasks.

That is why, when it comes to reliability, an agent-based approach can multiply potential issues by an order of magnitude if the underlying AI technique or practice isn’t up to reliability standards.

Conflation is the new operational risk

Operational trust starts with semantic discipline. Scale demands trust, not hype. While GenAI continues to evolve towards the reliability levels of foundational AI techniques, it shouldn’t stall other AI efforts already delivering value.

Semantics matter because they shape how organisations design, deploy and scale AI. Effectively leveraging AI isn’t about privileging one practice over another, but about mastering composite architectures and models. This means combining multiple AI techniques and anchoring emerging capabilities, such as GenAI and agentic AI, in the dependability of foundational AI and rigorous semantic control.

As Albert Camus famously observed, “Misnaming things is adding to the misery of the world.”

*Erick Brethenoux is a distinguished VP analyst and AI chief of research at Gartner. He is speaking at the Gartner IT Symposium/Xpo on the Gold Coast (14–16 September) on operationalising and scaling AI in the enterprise.

Image credit: iStock.com/akinbostanci