There is a great deal of noise around AI right now.
Some of it is justified. Large language models are genuinely impressive. They are already useful. They will, in time, make a radical difference to how we live, work and play.
But that does not mean the wider AI market is about to unfold at the speed many people now seem to assume.
If anything, the current moment reminds me of earlier technology waves where the technical case was real, the long-term direction was right, but the people, the market and the supporting conditions were nowhere near ready.
That distinction matters.
Digital transformation is a foundational step for organizations preparing for AI adoption, requiring them to benchmark their progress and achieve data excellence before moving forward.
Because markets do not move at the speed of demos. They move at the speed of trust, habit, incentives, infrastructure, commercial readiness and institutional willingness.
Cultural resistance to change, especially in industries with older workforces, is a significant barrier to AI adoption.
And those things move far more slowly than technologists like to believe.
Organizations must progress through digital and data maturity before they can fully embrace their AI journey, moving from initial readiness to actual AI model implementation and operational deployment.
The mistake of confusing capability with readiness
One of the oldest errors in technology is to assume that because something can be done, it will now be widely adopted.
That has almost never been true.
A technology can be technically sound, commercially interesting and directionally inevitable, and still take far longer to diffuse than its advocates expect. To drive successful AI adoption, organizations need a clear AI implementation plan and sufficient expertise to evaluate, deploy, and scale AI solutions.
I saw this with fibre-to-the-home over 20 years ago. At conferences back then, there were incredibly knowledgeable people who were already good to go. The vision was clear. The technical arguments were strong. The future was obvious to those in the room.
But the market was not ready. The people were not ready. And, crucially, the incumbents were not incentivised to move. Telcos had copper cash cows to sweat. Existing economics mattered more than technical elegance. In fact, a lack of identified need for AI and limited AI skills and expertise are the most common factors preventing businesses from adopting AI, with 71% and 60% respectively citing these issues.
By the time real FTTH became prevalent, many of those early pioneers had stepped aside and a new crowd arrived speaking as though they had invented the whole thing.
That is how technology markets often work. The early people are not necessarily wrong. They are often simply too early for the surrounding system.
AI has a very similar feel.
We have seen this before
The internet itself is a useful comparison.
People said it would transform everything. They were right. But it took time. More importantly, it took the supporting means to connect properly. Many businesses that were directionally correct arrived too early, when dial-up was still the reality.
The problem was not always what they were offering. The problem was that the conditions required for mass use had not yet arrived.
The same applies to AI now.
The models are powerful. But the surrounding readiness is weak.
Businesses still have messy processes. Data is fragmented. Ownership is unclear. Trust is low. Error tolerance is low. Buyers do not really know what to ask for. Employees do not want upheaval. Managers do not want risk. Existing vendors want to defend their position. Existing systems are entrenched. AI adoption also varies significantly across different business functions—such as information technology, finance, sales, HR, marketing, and legal—each of which requires tailored strategies to maximize performance and drive culture change.
So while the capability is there, the human and organisational readiness often is not. AI can automate routine tasks, allowing employees to focus on more strategic, high-value work and minimizing human error.
That matters more than many AI enthusiasts admit.
Why everything looks more ready than it really is
One reason the current AI moment feels more advanced than it actually is is because the people talking about it are often living inside a distorted environment.
Technologists are naturally drawn to what is possible. Venture capital is drawn to what looks investable. Platforms like X create a constant stream of demos, announcements, funding rounds, benchmark claims, launch threads and product clips that make the whole thing feel as though mass adoption is already underway.
It creates the impression that everything is good to go.
From inside that bubble, it can seem as though the world is racing toward autonomous agents, AI-native workflows and the wholesale rewiring of work. In reality, organizations often focus on specific AI applications that align with their strategic priorities, such as predictive maintenance, diagnostics, inventory optimization, or automation.
But that is not how most businesses experience reality.
Most businesses are not on X talking about model releases and agent frameworks. They are trying to keep the lights on, manage staff, retain customers, get invoices paid and deal with the drag of ordinary operations. The adoption of new tools is influenced by practical business needs and cultural attitudes toward technological change.
That gap matters.
The AI conversation online is often shaped by technically sophisticated people, investors who need scale narratives, founders who need attention, early adopters who enjoy rough edges, and a social media environment that rewards certainty, novelty and momentum.
The flow of VC money makes this even more powerful. Large funding rounds create the appearance of inevitability. If enough capital is chasing a space, people start to assume the market underneath must already exist at scale.
But that is not always true. In fact, the most commonly cited reason for adopting or expanding AI use among current and prospective AI users was to increase efficiency or productivity, reported by 65% of this group.
Capital can get ahead of adoption just as easily as technology can. In fact, that is often exactly what happens in major waves. Money, media and demos create a sense of arrival long before mainstream operational readiness has actually formed.
The real winner, for now, is the model layer
If I had to make the simplest argument, it is this:
The near-term winner is not the grand “agent economy.” It is the AI models and the simple AI tools and application layer around them that are driving current value.
Large language models are where the real value is today.
That value is not mainly in autonomous agent swarms or elaborate workflow narratives. It is in far more basic and immediately useful things: drafting, summarising, classifying, extracting, retrieving, coding assistance, natural language interaction with software, and reducing cognitive friction. Content creation is also a key application of AI tools, enabling businesses to generate marketing materials, emails, and visual content efficiently.
That is real utility.
Micro businesses in particular may go no further than this for quite some time. Many already have ChatGPT or Claude open, and for them that may be enough. They do not need elaborate agent frameworks. They need a faster way to write, think, search, summarise and occasionally tidy up some admin. Tools like AI-driven recommendation systems can provide hyper-personalized customer experiences that boost brand loyalty.
That is not trivial. It is useful. But it is also a long way from the louder narratives currently circulating.
It is also quite possible that, over time, smaller firms will run more of this capability locally on PCs or edge devices, only reaching into the cloud when they need real compute or scale. That would be a far more familiar pattern for many businesses: local where possible, cloud where necessary.
Again, the likely reality is more hybrid, incremental and boring than the more theatrical visions currently doing the rounds.
SME is too broad to be useful
One of the problems in this conversation is the lazy use of the term “SME.”
A sub-£100k owner-operator and a £10m business with an actual management team are not remotely in the same market.
Most micro businesses are not realistic buyers of serious AI services. They are more likely to use off-the-shelf tools themselves and stop there.
At the other end, larger SMEs and enterprises may play with AI, but many are still a long way from proving meaningful ROI. A great deal of enterprise activity today looks like experimentation, internal demos, sandboxing, workshops, pilots and innovation theatre.
That is not the same as embedded commercial value.
So where is the real market?
Mostly in the middle: firms with enough operational complexity to feel pain, enough digital maturity to absorb change, and enough budget to justify solving the problem properly.
But even there, the opportunity is more likely to be in practical workflow improvement and simple AI assistance than in the bigger stories people like to tell.
The slowness of people is always underestimated
One thing experience teaches you is never to underestimate how slowly people adopt new technologies.
This is not because they are stupid. It is because they have lives, habits, budgets, incentives and risks.
Most businesses are not trying to live on the frontier. They are trying to keep things running, avoid costly mistakes and preserve revenue. However, fear among employees—especially regarding job security, ethical concerns, and data security—can significantly slow AI adoption.
That creates a different rhythm of adoption than the AI world often assumes.
Many people in AI today are tech-first in their thinking. That clouds their view. They look at capability curves and imagine adoption curves will follow quickly.
But capability and adoption are different things. In fact, a study found that 29% of employees, including 44% of Gen Z workers, admit to sabotaging their company’s AI strategy due to job security concerns.
A demo can move overnight. An organisation cannot.
And social media makes this worse. When you spend enough time on X, surrounded by product people, VCs, model labs, founders and engineers all speaking as though the future is already here, it becomes easy to forget how little of that conversation maps onto the average buyer.
Online, a funding round looks like validation. A benchmark looks like commercial proof. A demo looks like product-market fit. A viral thread looks like market demand.
In the real economy, none of those things necessarily mean very much.
That is why I think many of today’s AI people are right about the destination but wrong about the speed.
The old Geoffrey Moore lesson still applies
Years ago there was a framework many technology companies used, associated with Geoffrey Moore: innovators, early adopters, early majority, late majority and laggards. Titles like Crossing the Chasm, Inside the Tornado and The Gorilla Game captured a truth that too many technology markets forget.
Early adopters are not the mainstream.
They tolerate rough edges, instability, changing standards and unclear ROI. Most businesses do not.
Most businesses want the opposite. They want proof, trust, support, familiarity and someone else to have gone first.
That is why I think AI is still much earlier in business adoption than the loudest voices suggest. While many are heading in the right direction with their AI strategies, the pace of adoption is slower than anticipated.
LLMs may cross first. Workflow adoption comes later. Agent adoption later still. Over half of organizations currently using AI (54%) feel ready to scale their use, while only 34% of those planning to adopt AI feel ready, highlighting a clear disparity in readiness levels.
The ability to do these things now may exist. But the people are not moving at the speed of early adopters, and they never do.
Distribution often beats innovation
Another historical lesson worth remembering is that the most exciting early movers do not always win the mass market.
Microsoft looked late to the internet party in the mid-to-late 1990s. But it had money, desktop distribution and control of a core layer. It could drop the browser into the operating system and change the game.
That is relevant to AI.
A great many startups are building clever AI features. Some will matter. Some will shape the category. But when markets harden, control of distribution often matters more than novelty. Software development choices—whether building AI solutions in-house, outsourcing development, or purchasing off-the-shelf tools—play a critical role in shaping how companies approach AI adoption and integrate these technologies into their operations.
That is why the larger platforms are so important. AI may not be adopted at scale first because people buy a new “AI product.” It may be adopted because it quietly appears inside the software they already use every day.
That is a very different kind of market dynamic than much current AI commentary assumes.
The same may happen again here. Today’s loudest AI-native companies may help define the category, but the broad commercial capture may still sit with those who already control the operating system, the productivity suite, the browser, the CRM, the design platform, the search layer or the enterprise relationship. AI adoption in businesses is also shifting from experimental pilots to operational integration, with organizations increasingly focused on achieving measurable ROI.
Generational memory matters
Another thing worth saying plainly is that many of the loudest people in AI today are under 40, and many are under 30. In many cases, they were not around when earlier waves of infrastructure enthusiasm, overbuild, adoption lag and post-hype disappointment were kicking off.
That does not make them wrong. But it can make them historically shallow.
If you have lived through the dot-com bubble, telecom overbuild, broadband delays, infrastructure arriving later than expected, and the way incumbents often absorb or crush pioneers, you tend to look at AI differently.
You know that a technology can be real and transformative while also being over-financed, over-sold and badly timed.
You know that being technically right is not enough.
You know that timing, distribution, incentives and market structure matter as much as the technology itself. In fact, 65% of organisations anticipate increasing their AI-related budget over time, showing a clear commitment to expanding AI capabilities.
And perhaps most importantly, you know that every generation eventually produces a crowd who say: this time it’s different. Yet, businesses continue to express hope for future AI advancements and easier access to off-the-shelf solutions as they look forward to the next phase of AI adoption.
Sometimes it is different in direction. But rarely in the speed, pain, casualties and consolidation required on the way there.
Data Security and AI
Data security is an essential pillar of any successful AI adoption journey. As organizations deploy AI systems that rely on vast amounts of sensitive data, the stakes for protecting that data have never been higher. AI technologies, including machine learning and natural language processing, are increasingly used to detect threats, monitor for anomalies, and respond to security incidents in real time. Yet, the adoption of AI also introduces new vulnerabilities—AI systems themselves can become targets for manipulation, data poisoning, or unauthorized access.
For businesses, this means that robust data security is not optional; it is foundational. Encryption, strict access controls, and regular updates to AI systems are now baseline requirements. Organizations must also ensure that their data governance practices keep pace with the evolving capabilities of AI, safeguarding both the data that powers AI and the outputs it generates. By making data security a core part of their AI adoption strategy, businesses can build trust, reduce risk, and create a secure environment for innovation. In the end, the success of any AI initiative depends as much on the security of its data as on the sophistication of its algorithms.
AI and Business Ethics
As artificial intelligence becomes woven into the fabric of business operations, ethical considerations are moving to the forefront. AI systems, if not carefully designed and monitored, can amplify existing biases or introduce new forms of discrimination—often in subtle, unintended ways. The use of AI raises important questions about accountability, transparency, and fairness, especially when decisions made by algorithms impact employees, customers, or broader society.
For organizations, addressing these concerns is not just a matter of compliance; it’s about building a culture of responsible AI use. This means actively working to identify and mitigate bias in data and models, ensuring that AI decision-making processes are transparent and explainable, and setting clear guidelines for how AI is used across the business. Employees need to understand not just how to use AI systems, but also the ethical boundaries within which they operate. By prioritizing business ethics in their AI adoption, companies can foster trust, reduce reputational risk, and ensure that their use of AI aligns with their core values. Ultimately, ethical AI is not just good practice—it’s essential for sustainable, long-term success.
AI Adoption in Different Industries
The rapid pace of AI adoption is reshaping industries across the board, but the journey looks very different depending on the sector. In healthcare, AI solutions are transforming diagnosis and treatment, helping clinicians make better decisions and improving patient outcomes. In finance, AI is at the heart of fraud detection, risk management, and portfolio optimization, enabling firms to respond to threats and opportunities faster than ever before. Manufacturing is leveraging AI for quality control, predictive maintenance, and supply chain optimization, driving efficiency and reducing costs. Meanwhile, the transportation industry is using AI to enhance safety, streamline logistics, and improve the passenger experience.
Despite these advances, each industry faces its own set of challenges—whether it’s regulatory hurdles, data privacy concerns, or the complexity of integrating AI into legacy systems. Understanding these unique challenges is key to developing AI solutions that deliver real value. The most successful organizations are those that take the time to understand their industry’s specific needs, risks, and opportunities, and tailor their AI adoption strategies accordingly. In the end, the right approach to AI is not one-size-fits-all, but one that is grounded in a deep understanding of the industry context.
The Importance of AI Education and Training
As AI technologies become more deeply embedded in the workplace, the need for comprehensive AI education and training has never been more urgent. For businesses, the real value of AI comes not just from the systems themselves, but from employees who know how to use them effectively. This means investing in training that covers everything from the basics of machine learning, natural language processing, and computer vision, to more advanced topics like AI ethics and strategic implementation.
Equipping employees with these skills is essential for driving innovation, improving efficiency, and ensuring that AI adoption delivers meaningful business results. It also helps address concerns about job displacement by empowering workers to adapt to new roles and responsibilities in an AI-driven environment. A well-trained workforce is better positioned to identify opportunities for AI integration, avoid common pitfalls, and contribute to a culture of continuous improvement. By making AI education and training a core part of their strategy, organizations can future-proof their workforce and ensure that their AI adoption journey is both successful and sustainable.
What I think happens next
My own view is fairly simple.
Large language models are the real winners for now.
Over the next five to ten years, the likely path is not a sudden autonomous agent economy. It is a slower bedding-in process.
First, the model winners establish themselves properly. Advanced AI capabilities now enable real-time analysis of vast datasets, supporting more effective and proactive decision-making.
Then those models become part of the normal software layer: documents, inboxes, browsers, operating systems, CRM systems, customer support tools, finance systems.
Then, after trust, standards, pricing, habits and workflows mature, businesses of all sizes begin to think more seriously about workflow automation and agentic models.
Yes, the capability exists now. But the people are not there yet. AI analyzes vast datasets in real-time, uncovering patterns and trends that guide strategic decision-making.
And people, in aggregate, do not move as quickly as the technologists.
They never have.
The bigger point
None of this is an argument against AI.
It is an argument against mistaking technical possibility for market readiness.
AI may well be as significant as the internet. But that does not mean the market around it forms overnight. There are a variety of artificial intelligence technologies—such as natural language processing, machine learning, and computer vision—being adopted at different rates across industries, with varying levels of awareness and application among businesses.
The history of technology suggests something much more uneven: real capability, exaggerated timelines, early casualties, incumbent resistance, infrastructure dependency, late mainstreaming and eventual normalisation.
And in the modern era, social platforms, venture capital and technology media can intensify that distortion, making every wave look more immediate, more inevitable and more commercially mature than it really is.
That is probably the more realistic way to think about the present moment. Around 1 in 6 businesses (16%) in the UK are currently using at least one AI technology, while 5% plan to adopt AI in the future, leaving 80% neither using nor planning to adopt AI.
The change is real.
But diffusion takes longer than excitement.
And that gap between what is possible and what people are ready for is where most of the real story lies.
