IT Strategy Matters: Gartner Top 10 Strategic Technology Trends for 2025 - What do they mean for your organisation?

Happy New Year! And welcome to the 8th edition of the IT Strategy Matters newsletter. I hope that 2025 is an enjoyable and successful year for you.  

I’ve put together a 10-day written course to delve into each trend that Gartner think will be important in 2025, and to help you to think through what each strategic trend means for your organisation. I’m sharing the first instalment of this course here in this newsletter to help you to get started on that journey.  

If you like this article, and want to sign up for the whole course, click here to access the rest of the course. 

Gartner is a globally recognised research and advisory firm that provides insights, advice, and tools for leaders in IT, finance, HR, customer service, and support. With decades of experience and a vast network of analysts, Gartner is renowned for its comprehensive research and strategic analysis. Their annual reports and predictions are highly anticipated and respected, offering valuable guidance to organisations navigating the ever-evolving technology landscape. 

Gartner has published 10 strategic technology trends that they think will be important in 2025. While not all trends will impact us immediately, Gartner believes we should start considering them in 2025. In some cases, proactive action is necessary to ensure they benefit rather than threaten our organisation.  

The trends that Gartner highlights are: 

AI imperatives and risks 

New frontiers of computing 

Human-machine synergy 

Of course, you can listen to what Gartner has to say about all of these, and I’m not intending to copy any of that. What I will do in this course is to delve into each trend, and to help you to think through what each strategic trend means for your organisation.  

Trend 1 – Agentic AI 

Unless you have been living in a cave for the past couple of years, there has been no escaping AI!  

AI isn’t new: The concept was created in 1956, but the capabilities of Generative AI that exploded into our awareness when OpenAI launched Chat GPT on 30th November 2022 have dominated much of the technology discussion since and are changing the way that all of us work. 

Artificial Intelligence (AI) has undergone significant transformations since its inception, evolving through various eras marked by technological advancements and shifts in research focus.  

The Dawn of AI (1950s-1960s) 

The concept of artificial intelligence began to take shape in the mid-20th century. Pioneers like Alan Turing and John McCarthy laid the groundwork for AI research. Turing introduced the idea of a machine that could simulate human intelligence, famously proposing the Turing Test as a measure of a machine's ability to exhibit intelligent behaviour. The Dartmouth Conference in 1956, organized by McCarthy, marked the official birth of AI as a field of study, where researchers began exploring problem-solving and symbolic reasoning. 

The Golden Years (1970s-1980s) 

The 1970s and 1980s are often referred to as the "Golden Years" of AI, characterised by the development of expert systems. These systems were designed to mimic human expertise in specific domains, such as medical diagnosis and financial forecasting. Notable examples include MYCIN, an expert system for diagnosing bacterial infections, and DENDRAL, used for chemical analysis. However, the limitations of these systems, including their inability to learn from new data, led to a decline in funding and interest, a period known as the "AI winter." 

The Resurgence of AI (1990s-2000s) 

The 1990s marked a resurgence in AI research, driven by advancements in computational power and the availability of large datasets. Machine learning, particularly neural networks, gained traction as researchers began to explore algorithms that could learn from data. The introduction of support vector machines and decision trees further expanded the toolkit available to AI practitioners. This era also saw the emergence of AI applications in industries such as finance, telecommunications, and robotics. 

The Age of Deep Learning (2010s) 

The 2010s ushered in the age of deep learning, a subset of machine learning that utilizes neural networks with many layers (deep neural networks). Breakthroughs in image and speech recognition, driven by large-scale datasets and powerful GPUs, led to significant advancements in AI capabilities. Notable milestones include the success of AlexNet in the ImageNet competition and the development of natural language processing models like Word2Vec.  

This era also saw the rise of AI in everyday applications, such as virtual assistants and recommendation systems, as well as the use of AI for Robotic Process Automation (RPA) to automate repetitive, rule-based tasks by mimicking human interactions with digital systems. 

The Era of Generative AI (2020s-Present) 

The current era of AI is defined by the emergence of Generative AI, which focuses on creating new content rather than merely analysing existing data. Technologies like Generative Adversarial Networks (GANs) and Large Language transformer Models (LLMs), such as OpenAI's GPT series, have enabled machines to both understand and generate natural language text, images, sound, music, and even video. This has profound implications for creativity, content creation, and human-computer interaction.  

Generative AI is transforming industries, from entertainment to education, and raising important ethical considerations regarding authenticity and misinformation. 

The main form that GenAI had taken so far has been as AI Virtual Assistants: Humans interact with the LLM through an application, traditionally a text interface, but now increasingly this is becoming multi-modal. The human gives the AI assistant an instruction through a prompt, and the AI goes away and quickly performs the task. Tasks range from research to large document summarisation, to taking meeting minutes, manipulating data or creating text, images or video.  

All of these capabilities help the human to be more efficient and effective, to be more creative or to do things that they previously did not have the knowledge or experience to do. In all cases though, the human is in control and the AI reacts to their commands.  

What these GenAI assistants did not do was to act autonomously in the way that we had become accustomed to with RPA. Whilst they represented a huge leap forward in capability, this came with a slower pace of work and less automation.  

I used to say that GenAI assistants like Microsoft’s Copilot are like a dog with a stick: you throw the stick. The dog quickly and skilfully retrieves the stick and brings it back. It then sits there patiently waiting to see what direction you will throw the stick in next.  

The Advent of Agents 

The pace of development and change of the underlying models, or LLMs has been staggering over the past couple of years. OpenAI alone has brought out a new model with mind-blowing new capabilities every few months, and each month the market has new entrants who are developing alternative models with different or more specialist capabilities.  

As we turn the page into 2025, the pace of model evolution seems to be slowing, but this is making way to an increasing pace of development with the applications that leverage the models’ capabilities.  

This is great for us as users and for our businesses because it means that use cases for this GenAI technology will multiply, and the tangible business value that it can deliver will be ever easier to realise. A lot of the commentary since GenAI models emerged has been ‘so what?’, ‘what do I use it for?’, or ‘how can I get tangible business value from these tools?’ which can require a significant commercial investment.  

The next wave of capability that we are seeing emerge is what is generally referred to as Agentic AI. This represents both an improvement in the model as well as improvements in the application layer to both deliver advanced reasoning and to perform tasks with more autonomy.  

Gartner thinks that this is a ‘now’ trend, with the main impact to organisations being seen over the next two to three years. Certainly, as we start 2025, there are many software vendors, like Microsoft and Salesforce, who are offering agents that promise to change the way that we work, and model developers like OpenAI are creating agentic models like GPT-o1 which are capable of advanced reasoning.  

OpenAI's GPT-o1 model delivers advanced reasoning through a combination of reinforcement learning and chain-of-thought reasoning. This model is designed to spend more time thinking through problems before responding, much like a person would. It refines its thinking process, tries different strategies, and recognizes mistakes to arrive at the most accurate and logical solution.  

This approach allows the model to perform at a level close to that of PhD students in areas like physics, chemistry, and biology. The model's ability to reason through complex tasks and solve harder problems than previous models is a significant advancement in AI capabilities. 

Agentic models 

LLMs and the GenAI assistants that use them are very knowledgeable generalists. Assistants like Copilot are designed to be able to answer many questions using general intelligence and large and wide-ranging sources of data. I like to think of Copilot like your brainy friend with whom you do the pub quiz: they have a dazzling level of information recall across a wide range of subjects. Enough to challenge the creativity of any quiz master and usually win the quiz. 

Such amazing generalists are usually not however deep specialists at anything. If you needed to have brain surgery, you are unlikely to ask your pub quiz friend to perform it. You would instead go to a brain surgeon who has spent years of their life studying and gaining experience in very specific knowledge and skills to that they can perform highly difficult but specialised tasks to a high level of quality.  

Agentic GenAI models change that. They are capable of advanced reasoning, and of creating highly accurate answers with consistently reliable results, because they break down a problem into specialist areas and leverage a combination of specialist agents to work on each one, with a specific set of knowledge. The model then brings together the responses from each of these agents to form the answer to the problem. Sometimes these agents work in parallel with another agent orchestrating the process, and sometimes they work in sequence, building on top of the response that previous agents have created.  

The result is that they can solve much harder problems with a higher degree of accuracy and reliability than previous models opening up ever more possibilities for their use.  

Autonomous Agents 

I’m going to have to stop using my ‘dog with a stick’ analogy. The application layer is developing to allow us to create agents that can work with us in a semi-autonomous way, or autonomously without us, carrying out their task automatically pro-actively or in response to a non-human trigger. 

These agents’ actions can be triggered by different events within digital systems. They can make decisions because they have memory, can plan, can learn from their mistakes, and can be linked to sensors and tools that allow them to engage with their digital, or maybe physical surroundings. 

This automation is much more than the RPA that we are used to. Rather than ‘robotically’ orchestrating a single process repeatedly, these agents can leverage the capabilities of the model, either classic or agentic, to be able to think, reason, learn and choose how to tackle a problem, rather than following one defines path and set of logic.  

With Microsoft Copilot Studio Agents for example, each business process can have different paths since agents create dynamic plans on the fly to handle and complete tasks. Users can view the underlying logic for each of the agents’ paths, which includes key details, steps, and systems involved. This provides insight into why the agent chose a particular method, its decision process, and context, along with detailed steps, including variables and outputs, which are crucial for debugging.  

Agents can stop and request human input, humans can review how an agent responded and train it to do so in a particular way in the future, or the agent can review its own work and revise how it might tackle a similar problem if it sees one again.  

As well as working autonomously, agents are the specialists that we were lacking in the world of GenAI. An agent can be trained to behave in a particular way. It can pre-prompt the model to provide a particular style of output or can be focused on a particular corpus of knowledge which can significantly increase the fidelity of the response that it provides compared to a more general AI assistant.  

I believe that both the use of both agentic models, and agents that we or software provider build will create a significant increase in the uses for GenAI and the value that it can deliver to our organisations. This could be the application of the capabilities which tips the scales from a business value perspective.  

Impact for you and your organisation 

So, what does this really mean for you and your organisation?  

Some of the answer lies in the details of what you will specifically use it for. Each industry and organisation have processes and data that are specific to them. This is often what differentiates you, and it could be that agentic AI can sprinkle even more magic on top of your current special sauce.  

There are of course, many use cases which are more common across organisations and industries. I’ll go through what some of these are to help you to think through how you could use them and what you need to do to see value delivered.  

First of all, a reminder of what sets Agentic AI apart: 

Gartner predicts that agents will make websites and applications redundant. Satya Nadella says that Copilot will become the ’UI for AI'. I don’t think that will happen any time soon, and the websites, intranets and applications are likely to still be needed to store data in a structured way and perform many business processes. I suspect at some point much of the traditional application layer will be transformed into something new, but to start with the GenAI assistants and Agents will be a new interface to the existing applications and data.  

I also think that there is an argument for retaining the current applications for digital inclusion: Some users will not use the AI as much as others or will be slower to learn how to use it most effectively and they will need access to the ‘legacy’ interface to be able to do their job.  

Here are some ideas of use cases for Agents that could be developed in a tool such as Microsoft Copilot Studio, or which might be developed and delivered by the software vendor (e.g. Microsoft or Salesforce): 

When I think about my own business at The IT Strategy Coach, my intention is to be a ‘solopreneur’. Whilst I may collaborate with many people, I hope to never have any employees other than AI ones. The possibility to scale a one-person business with today’s technology feels almost limitless, and a few years ago that was as far from the case as is possible!  

And I think that this approach applies to bigger businesses as well: if each of your employees will be able to scale to deliver an output significantly bigger that they could without the help of AI then your business can grow considerably. If some of that AI capability is agentic, that it will be like having a huge cohort of super-fast and super-effective employees on your payroll, but for a small fraction of the cost of the human ones.... 

But all is not lost for employees. Even with Agentic AI, there is still a need for human-led design, orchestration, quality control and most importantly, connection with other humans in the process. Fundamentally, the technology is there to help each and every one of us to achieve more, and to drive greater business performance as a result.  

In summary, I think that ‘AI agents’ will be great for: 

AI agents offer significant potential for enhancing efficiency and innovation. As we move forward, it is crucial to stay informed and proactive in leveraging these technologies to drive business success 

Uses for Agentic Models 

Remember, agentic models are capable of advanced reasoning, and of creating highly accurate answers with consistently reliable results, because they break down a problem into specialist areas and leverage a combination of specialist agents to work on each one, with a specific set of knowledge. The model then brings together the responses from each of these agents to form the answer to the problem. Sometimes these agents work in parallel with another agent orchestrating the process, and sometimes they work in sequence, building on top of the response that previous agents have created. 

Here are some potential use cases for an agentic model like Open AI’s GPT-o1 model:   

Things to be cautious about and prepare for 

Agents can be useful, but we also need to carefully govern the impact and implications of what they do: Ensure that you know when agents have been introduced into your software and processes and that the vendor doesn’t do this without telling you. And ensure that you understand the impact of using agentic AI and how it is regulated.  

Be cautious of BYOAI or BYOA: This is the next generation of Shadow IT. Never before have staff or leaders been able to engage in the latest technology revolution on a self-serve basis. There is great opportunity for IT to provide capability to staff and the organisation, but there is also great opportunity for staff to self-serve which could bring with it many risks.  

According to the Microsoft Work Trend Index report from May 2024, 75% of knowledge workers were already using AI at work and 78% of them were bringing their own AI tools to the office. That represents a significant data loss issue and potentially a privacy issue if users are sharing confidential information with these tools.  

Accuracy of data will be key. Many people have been saying this about GenAI assistants, and certainly misinformation or hallucinations have been an issue with these tools, but when a GenAI assistant is being used, there is always a human in the loop. It remains that human’s responsibility to check the output of the AI assistant and to ensure that the work produced is sensible and coherent, and hopefully accurate.  

If the agent is working autonomously, there is no human in the loop to verify this. We can do things like training the agent to check its own work, or in the case of agentic models, having one agent check the work of another, but the more accurate and well-structured the data that works from is, the more accurate the output from the agent will be. 

Be aware of the human connection and wary of dehumanisation: As AI takes over more tasks, there is a risk of employees feeling isolated and detached. Maintaining human connection and ensuring that AI complements rather than replaces human interactions is vital for preserving organisational culture and employee engagement 

Finally, regulation and ethics: there are some things that should not be decided by a machine! I’ll go on to cover this in more detail in the next lesson, but as is often the case with technology enhancements, the regulatory landscape is playing catch-up but is changing fast and it’s important that you are aware of your obligations and your risks. 

In the EU, AI decision-making was always a key part of the GDPR regulations. In addition, the EU AI Act which was approved by European countries and the EU Parliament in the first half of 2024, aims to create a framework for the development and use of AI in Europe. It focuses on ensuring that AI systems are reliable, robust, and trustworthy. 

In the US, In the 2024 legislative session, at least 45 states, Puerto Rico, the Virgin Islands, and Washington, D.C., introduced AI bills. Examples include Colorado's comprehensive AI legislation requiring developers and deployers of high-risk AI systems to use reasonable care to avoid algorithmic discrimination and disclose information to consumers. 

And of course, there is good old-fashioned information security. Beyond the risks that are posed by BYOAI, any AI tool that you use needs to conform to the same information security requirements as the rest of your IT infrastructure.