For the past few years, artificial intelligence has been framed as an unstoppable force, reshaping industries, accelerating productivity, and driving valuations to unprecedented levels. Now, that narrative is beginning to shift.
In a recent episode of XRaised, barrister and MiAI LAW CEO Laina Chan joined financial markets expert David Ioannidis to examine what many are calling a turning point for AI-driven businesses. The conversation did not center on whether AI will transform industries, that is already well established. Instead, it focused on a more immediate question: what happens when expectations outpace reality?
The answer, increasingly, is visible in the markets.
Recent developments, including the release of advanced contract review capabilities by major AI providers, triggered sharp reactions across the software sector. Billions in market value were erased in a matter of weeks. At first glance, the response appeared abrupt, even excessive.
But as Ioannidis explained, the shift was less about panic and more about correction.
Markets were not reacting to a single product release. They were adjusting to a broader realization that many AI-driven business models had been priced on assumptions that no longer hold.
For years, software companies, particularly those positioned around AI, were valued on the expectation of sustained, high-margin growth. Multiples expanded well beyond traditional benchmarks. The belief was that these platforms would continue to scale without meaningful disruption.
What recent developments have exposed is that this assumption was incomplete.
As core AI capabilities move upstream directly into foundational models, products that merely package those capabilities are becoming increasingly vulnerable. The distinction between a platform and a wrapper is no longer theoretical. It is now being priced into the market.
Chan pointed to a growing disconnect between public market corrections and private market optimism.
While listed software companies have experienced significant repricing, some privately held AI firms continue to command aggressive valuations, often without corresponding changes in revenue. This divergence raises an important question: how long can those valuations be sustained?
The answer, both speakers suggested, will depend on one factor above all others: defensibility.
In practical terms, this means asking whether a product offers something that cannot be easily replicated. If the value lies primarily in access to a language model or a simplified interface, the barrier to entry is low. As more advanced tools become available directly from model providers, that advantage erodes quickly.
On the other hand, products that integrate governance, accountability, proprietary data, or deeply embedded workflows may retain and even strengthen their position.
This distinction is particularly relevant in the legal sector.
Legal AI has emerged as one of the first verticals where these pressures are becoming visible.
Unlike many other industries, law operates within strict requirements around accuracy, traceability, and defensibility. Outputs cannot simply be plausible, they must be verifiable. Sources must be clear. Reasoning must be transparent.
This creates a higher threshold for adoption.
Chan noted that responses within the legal profession vary significantly. Smaller firms and regional practices often view AI tools as a way to enhance capability and compete more effectively. Larger firms, while actively exploring these technologies, approach adoption with greater scrutiny.
Barristers, in particular, have shown resistance, not necessarily because of skepticism about capability, but because of concerns around how these tools intersect with traditional billing models and professional identity.
This highlights a broader dynamic: technological adoption is not driven by capability alone. It is shaped by incentives, workflows, and deeply ingrained structures.
At the same time, the conversation challenged a common misconception about AI itself.
While large language models can process vast amounts of information and identify patterns at a scale beyond human capacity, they are not inherently reliable decision-makers. They generate outputs based on probability, not certainty.
This distinction matters.
Without clear sourcing, governance, and verification, AI-generated insights can introduce risk, particularly in fields where accuracy is non-negotiable. As Ioannidis noted, even sophisticated systems can produce confident but incorrect outputs, leading to flawed decisions if left unchecked.
This is where the concept of human oversight remains essential, not as a limitation, but as a safeguard.
Despite the turbulence, both speakers were clear on one point: this is not a collapse. It is a recalibration.
The term AI apocalypse has surfaced in some discussions, but it oversimplifies what is happening. Markets are not rejecting AI. They are becoming more selective.
The companies that survive this phase will not be those that rode the wave of early enthusiasm. They will be the ones that demonstrate real value through defensibility, reliability, and integration into meaningful workflows.
This pattern is not new.
As Chan observed, similar cycles have played out before. The early internet era saw a wave of rapid growth followed by consolidation, where only businesses with sustainable models endured. The same principles are now being applied to AI.
What makes this moment significant is not the correction itself, but what it reveals.
For years, the conversation around AI has been dominated by possibility. Now, it is being shaped by practicality.
The question is no longer whether AI can perform a task. It is whether that performance translates into durable value.
And in that shift, a clearer picture is emerging.
AI will continue to expand. Capability will continue to improve. But value will increasingly depend on what surrounds the technology, governance, trust, integration, and judgment.
In other words, the future of AI will not be defined by what models can do. It will be defined by what businesses can build around them.
