en
da
1. AI writes the code – and that changes everything for development speed
In February 2025, one of the biggest and sharpest AI minds, Andrej Karpathy, defined the concept of "Vibe Coding" in a post on X:
”There’s a new kind of coding I call “vibe coding”, where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. It’s possible because the LLMs are getting too good. I’m building a project or webapp, but it’s not really coding I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works.”
This description is wild – because it's not hype, but real. If a developer is still writing concrete code in 2026, something is wrong: AI can programme and does it well.
Funnily enough, AI's strength in coding is connected to its general weakness. Where AI always fails is when there's high ambiguity. Code is formal language. That means it's language defined precisely by very low ambiguity. There's nothing extra- or non-verbal. No hidden meanings. Everything is in the code. This means that large language models are almost tailor-made to work with coding solutions.
At the same time, it's also one of the use cases that has matured the most over the past couple of years. One of the major AI labs, Anthropic, has thus already built a good business on precisely creating AI models that are deeply specialised in coding.
Concretely, this means you can now expect: faster prototypes for development tasks and faster development of concrete new solutions and automation that can manipulate everything that can be manipulated with code.
Is it possible to do x or y, and does it pay off? In 2026, you'll be able to say yes to both far more often than before.
2. Transform AI from tool to workflow
A couple of years ago, the mantra was that everyone should have access to a raw AI language model. In Denmark, this has now become baseline. Now, the mantra instead is that AI must perform specific tasks that can be measured and validated.
This is also driving a general shift in the market away from simply having raw AI models. Instead, we're moving towards something that more closely resembles what we previously called software: purpose-built solutions that perform specific tasks.
And here's where it gets interesting: the powers that AI has to manipulate code (see first topic) also mean they can work for you and automate virtually everything that happens digitally – which is really quite a lot. Look away from this article and think about what's driven by, in and via digital systems in your company. The potential cannot be overlooked.
But as with all potential, it's not necessarily easy to fully realise. The winners will be the companies that are able to define three things: concrete tasks and processes, specific and accurate datasets and non-overlapping system landscapes for a well-instructed AI engine. Because, as with coding, ambiguity must be reduced.
And what provides the mandate for this work? Ownership, control and sharp and measurable performance expectations from AI.
In many ways, it's classic IT and project management and follows the almost hackneyed 'people, process, technology' prioritisation – but the possible gains are enormous when AI becomes software. At least if you do it properly and ensure solid frameworks for the projects you set in motion.
Top AI and change experts
3. Make your Copilot licenses productive – now
From time to time, when we at Basico have hosted AI events, we've asked the room how many have Copilot from Microsoft. Every time we see a sea of hands, so we can safely say that we're probably looking at 70-80% of all Danish companies that have AI through Copilot.
It's a great success – not least for the licence provider.
For whilst Microsoft has been busy adding the necessary security and control mechanisms to Copilot and rolling these out during autumn 2025, companies are already further ahead. They've started asking what use the many licences are actually doing.
Perhaps it can be compared to the rollout of Excel in the 90s – but otherwise it's completely and utterly unprecedented how big a task Danish companies have ahead of them in making the licences productive together with, by virtue of and despite their employees.
The fact is that far too few know why one should create specialised agents instead of raw language models. And presumably even fewer know how to subsequently create them. But it's precisely here – in the purpose-built, process-oriented agents – that the value in Copilot is found. It's concretely about getting AI into the process, so employees don't have to work out themselves how to take the process to AI.
Are agents difficult to build? An AI agent in Copilot can vary wildly in complexity, and whilst some become larger projects, others are easy and almost trivial to configure – but still with great gains. This is precisely where we're ready to advise you, so you focus your energy in the right places when you start building agents in Copilot.
Because the bottom line is that the licences in the vast majority of companies have already been purchased. With a Copilot licence at 150 kr./month per user, 100 relatively unused licences cost you 180,000 kr. a year. Do you have control over whether your investment is productive? The task in 2026 for Danish companies and for you as CFO is not to remove the licences again, but to make them productive, so that you're not only paying, but also reaping benefits continuously.
You can read more about how to get value from your Copilot licences here.
NewTech
4. Integrate AI into the work – not alongside it
And that brings us to the final point. Because there's a chasm between what AI can do and how it's used. Far too many workflows remain unaffected by what AI can help with – for several reasons:
- You've tried with a raw AI model that failed and have therefore concluded that AI cannot solve the task.
- You see AI as something separate, outside the tools you normally work with ‒ as a curiosity rather than a usable tool.
- Tasks must be moved over into an AI – there's no AI in the tasks.
Companies don't operate in isolation – they're embedded in complex ecosystems of systems and partners. And that's the core of the problem.
If this ecosystem isn't AI-ready – if your applications don't have built-in AI functionality – AI becomes an external discipline instead of a natural part of the work. The technology exists, but it needs to be integrated into the processes and tools where employees actually create value.
The challenge isn't technological capacity, but structural integration. As long as AI remains something you have to "go out to", instead of something that's built into the systems you use daily, the gap between potential and application will persist.
Fortunately, the task is understood – the vast majority of dialogues today are about how AI solves concrete and specific problems. Because once AI proves its worth and is set up properly and correctly, the ecosystem shifts quietly and steadily.
For three years with AI, the market has both feared a sudden and destructive upheaval of everything and simultaneously expected a positive and automatic transformation of the same. Both are misconceptions. AI can deliver impact, but it requires understanding what is being impacted.
Therefore, it's not just a good idea, but absolutely crucial to sit down, get serious and begin the hard work of being specific, so that the unlimited potential for value creation, productivity and new ways of solving tasks that AI contains can be released.
Happy working.
Do you want to know more about how your AI investments create real business value?
Then give our Lead AI & Technology Strategist, Lasse Rindom, a call. His engaging presentations will certainly leave you with AI food for thought. And are you looking for inspiration on how to use and implement generative AI in a way that aligns with your company's purpose, vision and strategy? Then he's also the person you're looking for.
