The AI Future We Are Promised May Not Be the One We Get

Why the future of work may depend less on replacing humans with AI, and more on knowing when and how to use it.

We are being promised an extraordinary future. One where artificial intelligence will perform most of the work humans do today, AI agents will manage entire business processes, and humanoid robots will take over much of the remaining physical work.

Perhaps this future will arrive. But the more I work with AI, both within my own business and with clients, the more I question whether this is the future most organisations will actually experience.

There is no denying AI's remarkable progress. What I question is the assumption that technological capability will automatically translate into widespread adoption.

There is a considerable difference between what AI can do, what organisations can reliably deploy, and what makes business sense to automate.

And that distinction is not receiving nearly enough attention.

Better productivity does not necessarily mean better business outcomes

For most professionals, generative AI may ultimately have an impact comparable to the introduction of Microsoft Office, albeit on a much greater scale.

Word processing, spreadsheets and presentation software transformed office work. Tasks that once required considerable manual effort became faster and easier, but people still needed to think, analyse, communicate and make decisions.

Generative AI is taking this further. It can draft documents, summarise information, analyse data, assist with coding and help people perform tasks that previously required specialised skills.

The productivity gains can be substantial, particularly for repetitive digital work. But productivity is not the same as effectiveness.

An employee who uses AI to produce ten reports instead of two may be more productive. Whether those reports lead to better decisions or improved business outcomes is another question.

And therein lies one of the challenges facing organisations investing heavily in AI: How much of the productivity gain actually translates into business value?

AI agents promise much more. But at what cost?

The next big promise is agentic AI: systems that can complete multiple connected tasks and potentially automate entire business processes.

Imagine an AI agent handling a customer enquiry, preparing a quotation, processing an order and arranging delivery, all with minimal human involvement.

The possibilities are exciting, and I would not rule them out. But making such systems work reliably involves far more than connecting a few AI tools.

Organisations need dependable data, clearly defined processes, integration across systems, exception handling, security controls and extensive testing. They must also maintain these systems as technologies and business requirements change.

Even an AI agent that achieves 99% accuracy at each of ten independent steps would have only about a 90% probability of completing all ten correctly. Actual outcomes depend on how errors interact and whether they can be detected and corrected.

A successful demonstration is very different from a dependable production system.

These challenges can be addressed, but doing so requires significant time, expertise and investment. The real cost of AI deployment extends well beyond subscriptions and computing resources.

Organisations are not simply buying AI. They are redesigning how work gets done.

Not every task needs generative AI

Another consideration is the consequence of getting things wrong.

Generative AI is remarkably versatile, particularly for tasks involving language, interpretation and multiple acceptable outcomes. But because it generates responses based on learned statistical patterns, its outputs are not guaranteed to be correct or consistent.

For drafting marketing content or brainstorming ideas, this may be acceptable. Humans can review and refine the results.

But what about calculating statutory payments, executing financial transactions or making decisions affecting people's livelihoods?

When errors carry serious consequences, organisations should consider more controlled approaches.

At FYT, we believe organisations need to recognise that generative AI is only one form of automation.

For tasks governed by clear rules, conventional programming or rules-based automation may be more suitable. For tasks involving predictions or pattern recognition, traditional machine learning may be appropriate. Generative AI offers advantages where flexibility, interpretation and multiple acceptable outcomes are important.

Sometimes, the best solution combines several approaches, including human oversight.

None is infallible. The question is which approach offers the most appropriate balance of reliability, cost and risk.

Growing regulatory requirements add to these considerations, particularly where AI is used in situations with serious consequences. Organisations remain accountable for their automated systems, regardless of the technology involved.

The question should therefore not be "Where can we use AI?" but rather, "What is the most appropriate way to get this work done?"

What happens when AI cannot access reliable information?

There is another emerging concern: the information that AI can access may be becoming increasingly selective.

One of generative AI's most attractive capabilities is helping users search for information, identify useful resources and build a case around a particular issue.

But this depends on access to reliable information.

Increasingly, website owners and publishers are restricting AI access to their content. A January 2026 study of 100 leading US and UK news websites found that 79% blocked at least one AI training crawler, while 71% blocked at least one crawler used for live AI searches.

These restrictions vary. Blocking AI training does not necessarily prevent live searches, and some AI providers obtain content through licensing arrangements.

Nevertheless, an important question remains: What happens when established publishers and other authoritative sources become harder for AI to access, while less reliable information remains readily available?

We cannot assume that less reputable websites are more accessible. But neither can we assume that the information available to AI represents a complete or balanced picture.

An AI-generated answer can be convincing, logical and supported by citations, yet still be misleading because important evidence was never considered.

This is not a new problem. In data analytics, even a well-executed analysis can produce misleading conclusions when the underlying information is incomplete or biased.

The difference is that generative AI often presents a polished, summarised answer, making it harder for users to recognise what might be missing.

This makes the ability to question sources, recognise gaps in evidence and challenge conclusions more important, not less.

So what happens to human work?

Much of the debate about AI focuses on which jobs will disappear. I believe a more useful starting point is to examine the tasks that make up those jobs.

Most jobs involve a combination of repetitive activities, analysis, communication, judgement and problem-solving. AI may automate some of these tasks without necessarily replacing the entire role.

Consider a financial analyst. AI may handle data preparation, reporting and preliminary analysis. But someone still needs to determine whether the right questions were asked, whether the data is reliable, whether important evidence is missing and whether the conclusions make business sense.

The same applies to HR, operations, marketing and many other professions.

As routine work becomes easier to automate, the remaining responsibilities may increasingly focus on defining problems, evaluating evidence, managing exceptions and making decisions under uncertainty.

This does not mean humans will always outperform AI. Human judgement can also be flawed. Rather, organisations need to understand where human judgement adds value, where analytics can strengthen it and where automation can reliably take over.

Even robotics and humanoid machines will face similar considerations. They already make business sense in certain manufacturing, logistics and operational settings. But broader adoption will depend on costs, reliability, safety and whether the benefits justify the investment.

Technology may expand what is possible, but business realities will determine what is practical.

The real competitive advantage is knowing how to combine capabilities

This brings me to what I believe will become one of the most important organisational capabilities in the age of AI.

For decades, organisations have invested in management, analytics and technology skills, often treating them as separate disciplines. But AI is making it increasingly important to bring these capabilities together.

An organisation with powerful AI tools but poor problem-solving capabilities may simply automate the wrong activities. One with sophisticated analytics but weak judgement may draw the wrong conclusions from accurate data. And one that relies entirely on human experience, without embracing analytics or automation, risks falling behind.

At FYT Consulting, we believe organisations need three complementary capabilities:

• Human Judgement and Critical Problem-Solving: Defining the real problem, questioning assumptions, understanding context and making decisions under uncertainty.

• Analytics and Evidence: Using reliable data to investigate problems, test assumptions, challenge conclusions and learn from outcomes.

• AI and Automation: Applying appropriate technologies to improve productivity, perform tasks consistently and scale what works.

But the real advantage comes from knowing when to apply each, how to combine them and when not to automate.

This is why FYT views critical problem-solving, data analytics and AI as parts of an integrated system of capabilities, rather than separate training subjects. Organisations can develop individual capabilities according to their needs, while building towards a more coherent approach to solving problems, making informed decisions and improving how work gets done.

Preparing for a future nobody can predict

I do not know whether the most ambitious predictions about AI will eventually come true. Neither, I would argue, does anyone else.

AI will continue advancing, and some applications that seem impractical today may become commonplace. But technological progress alone will not determine the future of work. Economics, regulation, reliability, access to quality information and organisational readiness will matter just as much.

My expectation is that AI will automate more individual tasks, selectively take over complete workflows and progressively reshape existing jobs. Some jobs will disappear, new ones will emerge, and many others will evolve.

The resulting workplace may be less autonomous than some AI companies envision, but potentially far more productive than the one we know today.

For organisations, the challenge is not to predict exactly how many jobs AI will replace. It is to develop the capabilities to adapt, make informed choices and continuously learn as technology evolves.

And for individuals, it is not simply about learning to use AI tools, but strengthening the judgement, analytical and problem-solving skills needed to work effectively alongside them.

The future may not belong to organisations that deploy the most AI. It may belong to those that make the best decisions about where, when and how to use it.

That is the future FYT is helping organisations prepare for.

Reference

Reuters Institute, University of Oxford — How many news websites block AI crawlers?