The greatest AI potential is not a chatbot

For businesses, the greatest opportunity in AI does not lie in chatbots, but in cheaper software that can close the gaps between systems. A genuine AI strategy is therefore about preparing the business for increasing complexity and customer demands – not about using AI for its own sake.

For almost four years now, the entire world has been preoccupied with the direct, chat-based AI experience. In 2026, however, the centre of gravity has begun to shift. As agentic AI coding matures, businesses are gradually shifting their focus from the direct use of AI to what AI can help them build.

The question is now less about what solution AI should be in itself and more about which solutions AI can help us create.

Three factors are particularly important in bringing about this shift:

• Code is purely digital. Code operates solely on what concretely exists in the code, on the screen and on the server.

• Code is a formal language. Programming languages are formal languages in which meaning does not depend on tone of voice, body language or hidden connotations. In many ways, code is language in its purest form, which naturally makes it well suited to a model that works with language.

• A large language model is, to put it somewhat tautologically, trained on enormous amounts of language. This naturally includes programming languages. The models know this language.

For AI – and for autonomous or automated processes in general – ambiguity is the crucial Achilles’ heel. When it is difficult to determine precisely which decision should be made, machines do not themselves possess the responsibility or judgement required to act or to ensure that they act in accordance with the intentions of their human owners. Nor should they.

However, as the three factors above show, this Achilles’ heel applies to a lesser extent to agentic AI coding. Coding simply contains so little ambiguity and is so closely linked to language that large language models can perform at a high level and be highly productive in this area.

This has several implications for modern businesses.

Cheaper code

First and foremost, it means that code will become cheaper. The cost of familiar solutions such as basic integrations or minor changes to existing user interfaces can be expected to fall significantly.

In the longer term, demand will become more complex because our ambitions will grow as it becomes cheaper and faster to develop new solutions. In the short term, however, while we optimise our existing solutions and work through years-old backlogs from before the explosion in agentic AI coding, cheaper code will set the agenda. Quite simply because most of our old problems are easy for AI to handle.

Cheaper code can close the gaps between systems

Looking a little further ahead, cheaper code also means that we can finally close the gaps between our best-of-breed systems – gaps that have so often led to headaches, emails, PowerPoint presentations, meetings, spreadsheets and so on.

When we implemented standard market solutions to reduce our investments and lower the cost of IT maintenance, we did not fully anticipate how much of our value chain would be lost or hidden in the gaps between the systems. It has turned out that a standard solution that everyone can buy is naturally not a complete answer to what creates a specific competitive advantage – such as yours.

This insight is neither new nor groundbreaking. Nevertheless, for some reason it continues to come as a surprise to many companies working with their enterprise architecture.

As agentic AI coding makes solutions cheaper, the gaps between standard systems can be closed more easily, without companies having to give up the benefits of SaaS solutions to the same extent. Your company may even be able to build lightweight applications that replace smaller SaaS solutions.

Does your company have thousands of spreadsheets in the salary adjustment process outside the HR and payroll system? Build an app and save many hours. It is as simple as that.

Knowledge loss can be reduced

This is where indirect AI also begins to meet the direct use of AI. By embedding language models in its applications, a company can analyse historical data in new ways.

What did we report in February, and how should that affect what we report at year-end? An AI-enabled reporting application can help keep track of the connection between the two.

In this context, AI addresses one of the central challenges facing every business, whether it is a global corporation, a medium-sized company or a scale-up: How do we reduce the loss of knowledge when employees leave, when there are too many projects for anyone to maintain a complete overview, or when legal risks grow across countries and over time?

An AI-enabled solution can, for example, identify that an incident in August is connected to something that happened as early as January. In this way, language models can keep track of what we have written in the past and bring historical knowledge into play when it becomes relevant again.

AI-enabled apps can help you keep track of what has happened and make it possible to respond more quickly and at the right time. At the same time, they can support more accurate reporting to clients, management, boards and investors.

This is also a concrete example of how important knowledge often exists outside core systems, in Excel spreadsheets, PowerPoint presentations and manual processes.

When we left the world of large monolithic systems and started buying standard systems, we gained agility, but we also lost something else.

Far too much negative attention and ridicule has been directed at the pitfalls of old monolithic systems, such as AS/400 systems and large customised solutions, while very few people have asked what they were actually good at.

Perhaps this is because standard systems gave advisors a good opportunity to challenge, rethink and transform old ways of working, while the large, heavy and expensive IT department provided a strong argument for change. Perhaps it is also because the systems gave new executives an opportunity to assert their authority and introduce new methods in their areas of responsibility, rather than simply acting as custodians of yesterday’s solutions.

Whatever the reason, it was not irrelevant detail that was lost in the enthusiasm for the new trends, and it deserves to be revisited. The old monolithic systems were built tightly around the companies’ core value chains – as a kind of early digital twin of the businesses they served.

What was lost was digital cohesion. This was subsequently patched together with vast quantities of spreadsheets in a Picasso-like enterprise landscape of overlapping systems that never quite fitted together. Which customer table is the correct one? The one in the CRM system? The ERP system? The e-commerce platform or the production system?

For bold companies, indirect AI and agentic AI coding represent an opportunity to revisit this and create greater clarity.

Perhaps you should not replace all your systems just yet. But if your most important differentiator is connected to your CRM system, you might want to examine whether your current system is tailored to your company’s needs. If your key differentiator is instead connected to how quickly you produce, for example, sofa cushions, you may want to take a closer look at your production system.

Perhaps the production system could even be built around and together with the e-commerce platform, so that the entire value chain is supported from end to end?

For most companies, it is not AI that is accelerating. It is the need to ask why that is arising more frequently and with greater urgency.

The ability of indirect AI to provide all kinds of solutions highlights and reinforces this need.

If cheaper code, better AI and closing gaps between systems are considered tactical areas of focus for the next three years, this raises the question: What should our strategy and strategic vision be?

The truly interesting thing about AI is not how quickly you can create something now, even though that has been the focus of this article. What is interesting is how much this speed will increase future expectations of what we can create and demand.

What will we need to do more of? How much more will our clients expect? And how much will this increase the complexity of our operations and value chains?

How do we prepare for this?

That question should be at the heart of your AI strategy. Because, in reality, it is a strategy for a world in which the ability to create and change solutions becomes significantly cheaper and faster.