AI can make software, but what does that mean for the industry?

On vibe coding and the future of software companies. I discuss AI in Dinero’s support, building an app and why trust and customers still matter.
Thorborg’s Diary · Episode 124 · 23 February 2026
Listen to the episode
Watch the episode on video
Listen in your podcast app:
Apple Podcasts – this episode · The podcast on Spotify · All episodes on Amazon Music · RSS Feed
Find more episodes here
Hello and welcome to Thorborg’s Diary. These are exciting times. Technology, society and AI are making waves. I’ve been in the media over the past few days, including Berlingske, and I can see some of what I’ve said is being understood a little differently from how I meant it.
Perhaps I’ve expressed myself unclearly too. But the headline about us eliminating thirty jobs certainly needs explaining. It’s both a little right and a little wrong, depending on the comparison.
AI handles questions, but not necessarily as much time
At the time of recording, AI handles around 71 per cent of questions in Dinero’s support. That doesn’t mean it’s removed 71 per cent of the work. AI starts with simple things, while the remaining questions are often more complicated. So you can’t simply multiply the number of questions into employee numbers.
I don’t have a precise figure for total time savings. My best guess is perhaps around forty per cent. But it’s an estimate. And it’s often the most repetitive tasks being removed. That’s something I like about AI: computers doing more of the boring work humans don’t need to sit with.
I know that can sound oblivious to privilege when I’m sitting in my lovely car saying it. If you liked those tasks and they paid your rent, their disappearance isn’t necessarily fantastic for you. We shouldn’t talk over that. I believe development can be good for humanity longer term, but that doesn’t make the transition easy for an individual.
We’ve had a hiring freeze for around eighteen months to two years while customer numbers grew around twenty per cent. This year we expect further growth. Without efficiency improvements, we’d therefore have needed more support staff. When discussing fewer positions, part of the difference is hires we haven’t made and won’t make.
The other part is that there actually is less work needing humans. Many support staff are students, including at CBS, and will move on to become accountants, consultants or something else after graduating. For example, seven or eight expect to leave this summer. We don’t need to refill those positions.
My starting point is that we can adapt largely through natural attrition. But I won’t give a guarantee I can’t stand behind. I hear leaders saying they’ll never dismiss anyone because of AI. I like the ambition to use the same people to do more, but I think the word never is dangerous. We don’t know exactly how work will develop.
Not everyone will want the new work
Some may say themselves: this isn’t what I signed up for. They liked their old task and don’t want the new one. If you enjoyed working with a shovel, you might not think sitting in a warm cab with a joystick for an excavator is fantastic. Even if others see it as a clear improvement.
The same can happen among programmers. Some see coding as a craft or art. They love building it themselves and don’t want to spend much of the day directing AI, checking suggestions and changing how they work.
Of course, there may still be a market for traditional craftsmanship. Just as people handcraft leather belts even though others make them more cheaply in a factory. But that won’t necessarily be how most of the market operates. Many of us need to adapt.
There are also people sneeringly asking whether the CEO escapes. No, of course I don’t. My work is changing enormously. I need to think differently about strategy, competitors, products, employees and organisation. And I can’t promise my own job security is unaffected either.
I’m 55. At some point, someone might look at me and ask whether I’m still the right person to lead a technology company. I think I’m keeping up well, and I’d like to continue for many years. But I’m not above the development I’m discussing.
AI also affects something I’ve practised for thousands of hours
I’ve spent many years becoming good at writing. I’ve read books, taken courses and practised for thousands of hours. Newsletters have let me measure what works. How many open? How many click through? Do the headline and text make people act?
I haven’t measured great literature. But I’ve learnt to explain something a little complicated so people want to keep reading and do something. That’s been an important skill for me.
I’ve written several books, some alone and some with others. I’ve collected much of my own material and given it to my AI as a basis for understanding how I write. Now ChatGPT can capture my points, tone and expressions pretty well.
And I have to admit that when I sit down with a text, it often does as well as I do. Sometimes better. And certainly much faster. That annoys me. It’s something I’ve invested a large part of my life in becoming good at.
I also had an experience with my good friend Søren, who wanted strategic advice about a business. Normally, I think that’s something I’m good at. I can see opportunities, analyse and help find a direction.
This time I suggested asking ChatGPT too. I gave it what I knew about the business and plans, and it produced an analysis. It contained much of what I’d have said myself, plus five or six excellent points I hadn’t thought of.
In that situation, I actually think AI’s help exceeded the analysis I could give. That was a bit of a bloody punch in the stomach. I’d imagined later in life sitting somewhere pleasant, perhaps abroad, helping business leaders as a coach or sounding board. That dream hasn’t disappeared, but looks different now.
Many companies might get much of their analytical help by learning to use a language model well. Then I need to be really good at something else too. Perhaps motivation. People tell me I can move them, give them courage and get them to act.
There’s a difference between teaching and a talk. Teaching should teach you something. A good talk should move you too. A twinkle in the eye, a pep talk and a loving kick up the arse can have value different from reading an analysis on a screen. I still think my experience, network and brainstorming ability can be useful. But I need to stay sharp. Just having been a CEO for years won’t be enough.
Vibe coding: Can you simply code a Dinero competitor?
The next question is highly relevant: if AI can build software, what does it mean for my own industry? I get comments saying someone can code something resembling Dineroin a weekend. Where does that leave us?
I think we need to distinguish between two ideas. Do you want to build a product and sell it to others? Or do you mean every small business should build its own accounting program in future? Both are worth discussing, but they have different challenges.
For example, my good friend Henrik Stendahl isn’t a developer. He’s good at culture and lots of other things, but isn’t the most technical guy. We have a freelance hairdresser who comes to Dinero, and employees pay her for haircuts. We just provide a room.
We needed a booking system, and Henrik built it himself with vibe coding in something like a couple of days. You can register, log in, see available slots and book. There’s also a reminder before the appointment. It’s bloody impressive that someone without a traditional technical background can make that.
It lowers the barrier to building small programs. And yes, it can create more competition. If I made my living from a single hairdresser booking system, I’d think hard about how strong my position really was. Of course, payments and other features can make a solution more comprehensive, but some basic functionality is becoming easier to build.
If fifty almost identical systems suddenly exist, choosing becomes hard too. Customers may choose the provider they know or the one others recommend. So a strong brand and trust may become even more important.
The interface isn’t the whole product
Take Mailchimp. We use it at Dinero too; we haven’t programmed our own copy to save the subscription. Could you build an interface for writing an email, selecting recipients and pressing send? Yes, you could.
But an email system’s value also includes the email actually arriving. When lots of emails suddenly come from a new sender or infrastructure, recipients’ systems may treat them as spam. Sender reputation, technical setup, abuse and delivery all need work.
You haven’t solved that just by building a nice send button. It’s ongoing work on operations and quality. Of course you can buy some infrastructure from others, but still need to understand what you’re building and who bears responsibility.
That’s what I mean by a moat around a software business. The code for the visible feature isn’t always the hardest thing to copy. Sometimes it’s everything around it that makes the feature work in practice.
Accounting requires operations, rules and security
Can you build an accounting program with AI? Yes, you can build a lot, and I think many are surprised by how many small details such a program contains. But let’s assume you can get the features built.
Then comes working with the rules. A standard digital bookkeeping system offered in Denmark must be registered with the Danish Business Authority. That requires compliance with the requirements. And it’s not something you do once and forget while the rules and product change.
If a business uses a custom-built, unregistered system, the situation differs: the business itself must ensure it meets relevant requirements. So your own system isn’t automatically prohibited, but responsibility doesn’t disappear because AI wrote the code.
Then there’s cybersecurity. If many businesses keep their accounts with you, protecting them is a serious task. An attack could block access or compromise data. Customers need to trust their accounts are stored and handled properly.
We spend many millions on cybersecurity at Dinero. I can’t determine a competitor’s security from its size. But I know it requires resources, skills and ongoing work for us. And there’s no guarantee any system can never be hit.
Then there’s trust. If a new business can choose between a hundred programs, it won’t necessarily pick the cheapest. It may choose the familiar name that’s been around a long time, has support and is expected to remain in a few years.
I’d think carefully myself before moving something so important to save a modest monthly amount. Others will choose differently, and that’s fine. But I don’t think price alone determines the whole market.
Integrations and customers are part of the moat too
Take a larger solution like Salesforce. A smaller business might manage with something simpler and cheaper. But a larger business may depend on many integrations and workflows. It’s not just one screen to replace.
We have many integration partners at Dinero too. Other companies have built their solutions against our API. That kind of ecosystem has value. It’s a bit like a phone: it may be technically fantastic, but if the apps you use aren’t available, something important is missing.
AI may make connecting systems in new ways easier. We’re following that too. But you still need to consider the whole package when assessing how easy replacing an established software business is.
So running a software business is about more than making software. Perhaps I can make something in my kitchen tasting roughly like Coca-Cola. That doesn’t mean I’ve built Coca-Cola. There’s production, distribution, brand, finances and customer relationships. The same basic principle applies here.
Acquiring a customer still costs money
We spend a lot on marketing. I can’t remember the exact current figure, but when we buy traffic, acquiring a paying user can cost several thousand kroner. And we do that as a known brand.
An unknown competitor might need to spend even more. Let’s use an example: acquiring a customer costs 5,000 kroner, and they pay 200 kroner monthly. That’s 2,400 kroner annual revenue. Before operations, wages and every other expense, it already takes around two years for revenue to cover marketing.
That’s not a precise forecast for every new accounting program. It’s an example of why cheaper code doesn’t automatically make building a business cheap. You may still need deep pockets and a long time before the finances work.
Looking at our paying customers at the time of recording, expected average revenue over a customer relationship is around 25,000 kroner. Some leave after a year, others stay a very long time. When we started, expectations were much lower. A new competitor doesn’t automatically have the same long history, retention or trust.
If you spend as much acquiring a customer as they’ll ever pay you, nothing remains for wages, rent or development. Those are things you need to solve beyond building the product.
Should every business build its own system?
The other idea is that a hairdresser, photographer, speaker or café can simply build its own accounting program. Perhaps getting an agent to build some of it will become easier. But why would that be a good use of their time?
They need to run their business. They need customers and to deliver their work. Should they also handle backups, rule changes, bug fixes, mobile versions and adjustments when a new browser or phone version changes something? All to save a subscription?
At Dinero, we take the opposite approach: if we can buy software doing around eighty per cent of what we need, we normally do. We haven’t built everything ourselves. We buy components for scanning, payments, subscriptions and email delivery, among other things.
We have plenty of developers. But that doesn’t mean assigning them to copy everything we pay others for. A mobile scanning component can cost us a couple of hundred thousand kroner annually. In return, others ensure it keeps up with iOS and Android, is maintained and gets new capabilities.
Compared with dedicating our own people to building and maintaining it, that’s often a very good deal. That remains true if AI makes development cheaper. Maintenance and responsibility are still work.
I can also wash the car, clean and trim the hedge myself. I don’t necessarily do it. I can spend that time on my business or family, and both can be worth more to me. Just because you can do something doesn’t mean you should.
I’m not talking down AI programming
This isn’t an attempt to say vibe coding is irrelevant. Quite the opposite. We already work with it and programming agents at Dinero. At the time of recording, my CTO estimates we’re around thirty per cent more productive without working harder. That’s insanely good.
I think it’ll revolutionise development. And some software businesses will face enormous pressure. Their products are easy to replace, their moat is small and customers can get a good alternative far more cheaply.
Others have more things that are harder to copy: rules, security, integrations, operations, trust, customer relationships and brand. So I don’t think every software subscription business will have exactly the same fate.
I also follow the large software companies and the turmoil around their valuations. But a falling share price doesn’t itself tell you which products survive. You need to examine where earnings come from and what specific advantages they have. Valuations could fall generally while some companies do far better than others.
We need to use the development before it uses us
At Dinero, we’re working flat out on our AI adviser. We have a team focused on those opportunities while also using the tools internally. It needs both to improve the product for customers and make us more efficient.
There are around 110 of us at the time of recording. If we’d followed the old growth curve without changing how we work, we’d have been substantially more. Now we can serve more customers with the same organisation, and I also expect fewer support staff.
Meanwhile, our industry has many data, security and bookkeeping requirements. We need to get that right. It’s not something I can hand off and forget. Management must take responsibility for protecting customer information and complying with the rules applying to us.
I certainly think AI can affect Dinero too. I’m not imagining we’re untouchable. But I believe we have a stronger moat than many others because we’ve worked on the product for a long time, have a strong brand and talented people and resources to develop further.
Am I nervous? I’m certainly a little paranoid. I’m actually fine with that. The day you get smug and say it could never happen to you, you’re heading for the grave. I need to keep asking what’s changing and how we respond.
I spend three or four hours a day thinking about AI, the future and society because it fascinates me enormously. That’s why I keep talking about it. Have a fantastic day.