100 Blogs Later: What Mercia AI Has Learned About AI

So, here we are, 100 blogs for Mercia AI.

Most of the time, the next article simply came from a question that seemed worth exploring. Sometimes that was a major AI announcement. Sometimes it was something affecting small businesses. Sometimes it was an experiment I wanted to try. And increasingly, it was something happening in sport that offered a surprisingly useful way to understand artificial intelligence.

Somewhere along the way, those questions became 100 published blogs.

At 50 blogs, I paused for a similar look back. Fifty posts later, the questions had changed again.

Reaching that number feels like a good excuse to look back — not simply at what was written, but at what changed along the way.

Some topics performed exactly as I might have expected. Others absolutely did not.

Some ideas became recurring themes, while others disappeared almost as quickly as the AI news cycle that produced them. And on a few occasions, writing about AI was no longer enough. I wanted to go beyond explanation and actually test what I was writing about.

The Questions Became More Interesting

Some of the earliest Mercia AI blogs followed the rapid changes taking place across the AI industry.

That was understandable. In early 2025, there seemed to be a new model, company, product or claim appearing almost every week. There was always something new to write about, and the pace of change made it feel important to keep up.

But after a while, simply reporting what had happened became less interesting than asking what it actually meant. The focus gradually shifted from “what is new” to “why does this matter”, and that change shaped the direction of the blog more than anything else.

That is when the questions started to deepen. What does the enormous growth of AI infrastructure mean for the resources needed to support it? Can Europe realistically build enough AI capability to compete with the United States and China? Who should be responsible when an AI system produces a harmful or incorrect result? And beneath the apparently digital world of artificial intelligence, what physical infrastructure is actually keeping everything running?

That shift produced articles such as Will Europe’s AI Ambitions Succeed?, which became part of a wider series exploring Europe’s position in the global AI race. Looking back, the subject feels even more relevant now than when I first wrote about it. AI sovereignty, computing capacity, energy and infrastructure have moved from fairly specialist discussions into much broader political and economic debates.

Another early example was Thirsty Machines: How AI’s Water Consumption is Reshaping Data Centres. AI can feel strangely weightless when you use it. You type something into a box and an answer appears a few seconds later, almost as if nothing physical is happening at all. But behind that interaction is an enormous system of servers, electricity, cooling equipment, specialist hardware and water usage that is easy to overlook.

What surprised me most was not just the topic itself, but the response it received. Thirsty Machines became one of the most-read articles on the Mercia AI site, and that was an early reminder that the biggest AI story is not always the newest model or the most impressive demo. Sometimes people are more interested in what is happening underneath the technology they are already using every day.

I Would Not Have Predicted Our Biggest Blog

If you had asked me early on which Mercia AI article would eventually become the most-read on the site, I am not sure I would have picked local AI inference.

But that is exactly what happened.

What Is Local AI Inference — and Why It Might Change How You Use AI has attracted more than 340 views and over 1,100 Google Search impressions, making it our strongest-performing article so far.

On the surface, the subject is relatively technical. It is about running AI models on your own hardware rather than relying entirely on cloud-based systems. It is not the kind of topic you would automatically assume would reach a wide audience.

Yet in hindsight, its popularity makes more sense. As people become more familiar with AI, their questions naturally become more practical and more personal. They start to ask whether they really need the largest model for every task, whether their data has to leave their device, and whether there are alternatives that offer more control or lower cost.

Those questions led to further articles, including Is Local AI Still Worth It for Small Businesses in 2026?, and they also changed some of my own thinking about how AI should be used in practice.

It became increasingly clear that there is unlikely to be a single “correct” answer to where AI should run. In some cases, cloud systems are clearly the right choice because of scale and capability. In others, local models are more appropriate because of privacy, cost or simplicity. In many real-world situations, the most sensible approach is likely to be a combination of both.

The lesson is not particularly dramatic, but it is important. It is less about one technology replacing another, and more about choosing the right tool for the specific job in front of you.

My Thinking About AI and Work Changed Too

The relationship between AI and employment has been another recurring theme throughout the blog.

Early discussions often seemed to fall into a very simple binary. Either AI will replace everyone, or it will replace nobody. Over time, neither of those positions felt particularly useful or realistic.

As the technology developed, the more interesting questions shifted towards tasks, workflows and how people actually interact with systems in practice. That is where the real change seems to be happening, not in total job replacement, but in how work is structured and supported.

The Rise of the Frontier Firm: What It Means for You and Your Work explored the idea that organisations may increasingly combine human employees, AI assistants and autonomous agents within the same workflow. Other articles looked at productivity, job searching, AI agents and automation, each from slightly different angles but with a similar underlying focus.

Writing about these topics repeatedly reinforced something that has become increasingly important to me. AI can be extremely useful, but it can also be extremely convincing when it is wrong. Those two things are not opposites, and they can exist at the same time within the same system.

Because of that, I have gradually become less interested in the question of whether AI can do something at all, and more interested in whether it can do it reliably enough to support better decision-making. That shift in focus now runs through a lot of what Mercia AI publishes, even when it is not stated explicitly.

Then Sport Took Over More of the Blog Than Expected

AI in sport did not begin as an attempt to turn Mercia AI into a sports-focused publication. It started more simply than that, because sport kept providing clear, real-world examples of AI being used in ways that were easy to observe and explain.

Over time, readers also responded strongly to those examples, which naturally encouraged more of them.

How Formula 1 Uses AI: We Are Checking became one of the most-read articles on the site, and The Rise of AI in Football: From Champions League Finals to Grassroots Strategy also performed strongly. From there, the coverage expanded into rugby, cricket, tennis, American football, golf and major international sporting events.

Eventually, it became clear that sport was doing something particularly useful for Mercia AI. It was making complex AI ideas easier to understand by anchoring them in familiar situations.

A Formula 1 race strategy, for example, can demonstrate how decisions are made under time pressure using constantly changing data. Football analytics can show how individual data points are combined into broader patterns. Broadcasting systems can illustrate computer vision, automation, translation and recommendation systems working together at scale. And prediction, perhaps more than anything else, highlights something fundamental about AI that is easy to forget: uncertainty.

The sport may change, but the underlying lesson often extends far beyond it.

Eventually, Writing About AI Was Not Enough

At some point, I became curious about something slightly different. Rather than only asking how organisations were using AI in Formula 1, I wanted to see what would happen if AI systems were asked to analyse Formula 1 directly.

That led to Lights Out and Away the LLMs Go! Can AI Predict the 2025 F1 Champion?, where multiple AI models were given the same challenge and asked to produce predictions based on available information.

What became interesting was not just the final answers, but the reasoning behind them. Which factors did each model prioritise? What did they ignore? How did they balance recent performance against historical data? And perhaps most importantly, how confident did they sound when dealing with something that is inherently uncertain?

Later, How Four AIs Predicted the 2025 F1 Champion — And Why the Margin Matters allowed those predictions to be compared against real outcomes, which added another layer of insight. Then F1 2026: When Four AI Models Agree in a Reset Year explored what happens when the underlying rules of the sport change, and historical data becomes less reliable.

That last point is particularly important. AI systems are very good at finding patterns, but they are much less reliable when those patterns stop holding. That is not just a Formula 1 problem. It is something that appears in business, economics and everyday decision-making whenever conditions change faster than historical data can keep up.

Then We Tried It With the World Cup

Formula 1 was not the only experiment of this kind. Ahead of the 2026 World Cup, multiple AI systems were asked to predict the tournament winner, which became Can AI Predict the 2026 World Cup Winner? and went on to become one of Mercia AI’s most-read posts of 2026.

As with Formula 1, the most interesting part was not the prediction itself, but the reasoning behind it. Different AI systems, given similar information, could still reach different conclusions, emphasise different factors, and express different levels of confidence.

That is an important reminder whenever AI is used for forecasting. A prediction can look precise, structured and confident, but that does not make the future itself any more certain. The appearance of accuracy and the reality of uncertainty are not the same thing.

Over time, these experiments became one of the most useful ways of exploring AI, because they made both its strengths and its limitations visible at the same time. AI can process information quickly, identify patterns and compare scenarios, but it cannot remove uncertainty from the world it is analysing.

The World Cup Became Bigger Than One Prediction

The 2026 FIFA World Cup eventually developed into a broader series of AI-related articles, rather than a single prediction exercise.

How AI Helps Broadcast the 2026 FIFA World Cup explored how technology supports the delivery of a global sporting event, from camera selection and highlights generation to translation, metadata and content discovery. These are systems that most viewers never directly see, but which play a significant role in shaping the experience.

Other articles looked at the wider infrastructure and complexity involved in running an event of that scale. That process reinforced another useful lesson: sometimes the best way to understand AI is not to start with AI itself, but with a real-world problem and then examine where the technology fits into it.

Mercia Minds Taught Me Something Different

One of the most surprising results in the site analytics had very little to do with algorithms or models. Building a Community Takes Time – Mercia Minds Hits the Road became one of the most-read posts on the entire site.

Mercia Minds was an attempt to take conversations about artificial intelligence out of the digital space and into real-world community settings. It was an experiment in a different kind of engagement, and it quickly revealed something that is not always obvious from online work alone.

Building something offline is fundamentally different from publishing online content. A blog can be written, edited and published in a relatively controlled process. A community, by contrast, has to develop over time. It depends on people showing up, on timing, on location, and on communication that cannot be fully automated or predicted.

Not every attempt worked as expected, and some ideas needed to be adjusted or reconsidered along the way. But those moments were often just as valuable as the successful ones, because they revealed what people actually responded to in practice rather than in theory.

That experience also reflects something more general about running a small business. Progress rarely looks as smooth or linear from the inside as it does in hindsight.

100 Blogs Changed Mercia AI Too

Looking back across 100 articles, one thing becomes clear. The blog did not simply document Mercia AI; it helped shape it.

Writing repeatedly about local AI made topics like privacy, cloud versus local processing and hybrid systems more central to the overall direction. Writing about sport made it easier to explain complex ideas through familiar examples. Testing AI models on real-world predictions brought uncertainty and reasoning into sharper focus. And writing about work and small businesses kept bringing everything back to a simple question: whether any of this is actually useful.

That question matters more than it might initially seem, because AI makes it very easy to produce large amounts of content very quickly. The challenge is not production, but usefulness. More output does not automatically mean more value.

After 100 blogs, that distinction has become more important than it was at the beginning.

What Have I Learned After 100 Blogs?

The technology has changed significantly since the first articles were published. Some tools that once felt central now feel like part of an earlier phase, while others have become normal far more quickly than expected.

Despite that pace of change, a few lessons have remained consistent. The largest model is not always the best model for a given task. A confident answer is not the same as a correct one. Data quality, privacy, cost, infrastructure and context all matter. And perhaps most importantly, human judgement still plays a central role in deciding how AI should be used.

Over time, it has also become clear that AI is easier to understand when it is applied to real situations rather than discussed in abstract terms. Whether that is a business problem, a sporting event or a local experiment, the value tends to come from seeing how the technology behaves in context.

That leads to the question that now sits underneath much of the Mercia AI blogs: how can AI help people make better decisions without overstating what it is actually capable of?

After 100 blogs, that question still does not feel fully answered. Which is probably a good thing, because it means there is still more to explore. And explore Mercia AI will.