The Reality of AI Today

AI finds itself at a strange impasse.

In some circles, it is still being hailed as the future, driving unprecedented levels of infrastructure and financial investment. In others, it is increasingly being viewed through a lens of realism, with the novelty beginning to fade and important questions emerging around value and sustainability.

This shift is occurring because organisations, and even individuals, are beginning to move beyond what AI could become and are asking harder questions about what it actually delivers today.

Three factors provide a more grounded perspective on the technology. As the market matures, they are becoming increasingly important in determining AI’s role in the future.

LLMs Are Not Synonymous with AI

One of the most persistent misconceptions is the conflation of Large Language Models (LLMs) with Artificial Intelligence itself.

AI is a broad field encompassing many technologies and disciplines. LLMs are one implementation within that field, yet much of today’s discussion treats the two as synonymous.

This distinction matters because LLMs, like all technologies, have inherent limitations. At their core, they are probabilistic systems that generate outputs based on patterns learned from their training data. They can be extraordinarily useful, but they remain dependent on reliable data, human input, oversight and correction.

This is very different from the idea of “AI” that often exists in the collective imagination. Unfortunately, marketing has done little to separate perception from reality. In many cases, it has reinforced narratives that imply a direct path from today’s LLMs to Artificial General Intelligence (AGI), despite there currently being no proven or universally accepted path connecting the two.

As a result, expectations have often become disconnected from actual capabilities, inevitably leading to disappointment.

The Looming AI Bubble

The second challenge is economic.

Much of today’s AI capability has been built on enormous investment and expenditure. Vast amounts of capital have been committed to model development, data centres, energy capacity and supporting technologies, based largely on seemingly unlimited expectations of future capability, demand, growth, and returns.

This investment has enabled many AI services to benefit from substantial capacity and subsidised pricing as providers compete to accelerate adoption and build market share. However, as the true costs of infrastructure, compute, energy, and model development become more visible, questions naturally emerge about sustainability, particularly when those costs are considered alongside the technology’s practical limitations.

Financial relationships within the AI industry are also becoming increasingly interconnected. Major technology providers are investing in, financing, and purchasing services from other participants across the same ecosystem, creating increasingly circular financial relationships. While this can accelerate development, it can also make underlying economics and genuine demand more difficult to assess, potentially increasing exposure to broader risks.

External pressures add further uncertainty. Economic conditions, infrastructure constraints, opposition to new data centre developments, energy requirements, and changing investor expectations are exposing potential weaknesses in the economics supporting AI.

These conditions make several questions increasingly relevant:

• What will AI services actually cost?
• Can model and infrastructure development continue at the current pace if available capital declines?
• What happens if a major AI vendor can no longer maintain today’s level of service?

Together, these factors create uncertainty around whether the current trajectory of AI is financially and operationally sustainable.

Even organisations that are not directly investing in AI should pay attention. The interconnected nature of the technology industry creates broader financial and operational risks. A significant disruption affecting a major AI provider, infrastructure operator, cloud platform, or investor could have consequences for other technology services, such as cloud software licensing, as well as industries beyond technology.

Productivity Is Not the Same as Value

Perhaps the most important question is not whether AI improves productivity or how much it costs, but whether it creates meaningful value.

This is where many AI discussions become blurred.

As AI’s honeymoon period begins to fade, organisations are looking more critically at the results of early adoption. Novelty and experimentation have undoubtedly driven widespread uptake, but novelty is not the same as value, and usage is not the same as a sustainable business case.

Productivity is frequently cited as AI’s primary benefit. If AI can summarise information or generate a document template faster, it may appear that productivity has increased. However, productivity is notoriously difficult to measure, particularly in knowledge-based work. The challenge becomes even greater with LLMs, where outputs often require continued human input, review, correction, and oversight.

More importantly, increased productivity does not automatically translate into business value.

Consider a baker who invests in a machine capable of producing ten times more bread.

On paper, productivity has improved dramatically. However, if customers no longer consider the bread to be as good as the original product, they may leave. Alternatively, there may never have been enough demand to justify producing ten times more bread in the first place.

In both situations, productivity increased, but value did not.

The same principle applies to AI. Similar challenges arise when measuring improvements in quality or creativity because of the inherent limitations of LLMs.

Generating more content, producing more ideas, or completing tasks faster may sound impressive. The real question is whether those gains produce measurable business outcomes that justify the investment and associated risks.

Organisations must also account for the governance, architecture, security, training, and operational requirements involved in implementing and managing AI. These activities create additional overhead and place further demands on already constrained IT and business resources.

The value calculation must therefore extend beyond licence costs or time saved. It should include the costs of implementation, oversight, correction, governance, ongoing management, and the risks associated with AI.

AI as a Personal Assistant

While AI can seem like a complex concept, there is a straightforward practical analogy.

Imagine hiring a human personal assistant for every employee in your organisation. The goal is to make each employee more productive by transferring routine and seemingly unnecessary parts of their workload to the assistant.

You soon discover, however, that the assistants cannot reliably perform every task originally planned. Differences in skills, responsibilities, and workflows mean employees must regularly explain tasks, review outputs, correct mistakes, and maintain oversight, effectively creating a tax on their productivity.

At the same time, uncertainty remains around the future cost of these assistants. They were initially hired at a substantially reduced price, but there are continuing signs that costs may increase, capacity may become limited, or the assistants may be unable to take on additional workloads.

Meanwhile, the organisation continues to carry the overhead associated with onboarding, governing, supporting, and managing this new workforce.

In many ways, this is AI in a nutshell. The assistants may still provide genuine value, but only when their capabilities, limitations, management requirements, and total costs are properly understood.

The Key Takeaway

LLMs and other modern advancements in AI have demonstrated impressive technological capabilities. However, the discussion is increasingly shifting away from what the technology can become and towards the value it creates today, whether its current trajectory is sustainable, and how much organisations should depend on it.

The takeaway is not that today’s technology has no place within an organisation’s technology suite. Rather, its role should be determined through a clear understanding of three interconnected factors:

• Capability: What can the technology reliably deliver today?
• Sustainability: Are its costs, capacity, and supporting economics sustainable?
• Value: Do the outcomes justify the investment, risk, oversight, and dependency involved?

Together, these factors provide a more grounded basis for determining whether AI creates sustainable business value and how much an organisation should invest in and depend on it today.