A glimpse of AI in 2026 reveals a mature technology, but still a lot of bullshit bingo. With all the bold predictions, the key skill is to stay flexible.

2025 was a wild ride, in many respects. In this article, let’s pick one specific topic and review the year 2025 and dare an outlook into 2026: Artificial Intelligence.

Ever since the launch of ChatGPT in 2022, I was neither fully sceptical nor fully euphoric about AI. As a sober engineer and tech entrepreneur living in reality, I have watched the topic critically, yet with interest.

Here is how I see AI at the turn from 2025 to 2026.

The Technology Has Matured

Let’s be clear: AI technology has matured to the extent that it has definitely changed our lives for the better.

For everyday use, language translation is my favorite example. Whilst early AI-assisted translation tools were inaccurate and clumsy, we’re nearing the point where you can have a phone conversation in your mother tongue with somebody speaking to you in Swahili, Sanskrit, or any other exotic language. The AI can handle real-time translation and make your counterpart appear to be speaking to you in your mother tongue. That’s how far the technology has come.

While Google Search was the standard for searching and finding information on the internet for decades, it is slowly being replaced by AI search tools. Already today, many people use ChatGPT or other AI chatbots to ask complex questions that would have needed many parallel Google searches in the past.

In my daily work as Founder & CEO of Yonder, a B2B SaaS company, I’m using various AI tools to prototype features and feature improvements. I can interact with those tools in natural language, and I get a presentable click-dummy I can show to a customer, and then to our development team. Product stories created using AI prototyping tools are significantly better than the legacy text-based product stories that always started with “as a user, I want…”.

As a last example, let’s look at converting data from one format into another. I’m not thinking of easy tasks such as creating a PDF file from a Microsoft Word file, or exporting a JPEG picture as a PNG. In our company, we often have to support our customers with importing legacy documents into our documentation software. Legacy documents can be aircraft manuals edited with PDF overlays, regulations that were originally scanned from paper, or tables and mathematical formulae inserted as screenshots instead of properly structured data. Try to do any of those tasks manually (we know what we’re talking about; we’ve been around longer than AI tools exist). And enjoy the moment you realize how much manual work contemporary AI tools eliminate for such tedious tasks.

AI Bullshit Bingo Is Still Going On

Even though the technology has doubtlessly matured, there is still a lot of AI bullshit bingo going on.

Most marketing departments still shout “AI! AI! AI!” all day long on all channels. Whilst that might catch some leads, some markets are still sceptical about AI, and for many use cases, you really don’t need to incorporate AI into your software product. Rather, you should use AI technology as a — very powerful — tool to assist your customers.

If you leave the B2B SaaS space and consider AI in the B2C domain, there are still lots of people who claim they can build a scalable software using vibe coding tools in an afternoon. As described above, vibe coding tools are great for prototyping, but to create scalable software solutions, you still need software engineers who know what they are doing. Maybe they’re more productive thanks to AI coding tools than they were in the past, but you still can’t create scalable solutions without knowledgeable software engineers.

Last but not least, medium.com and other platforms are full of articles from people who claim they got rich thanks to AI tools that helped them build billion-dollar companies just by themselves. I consider that to be complete nonsense.

AI Unit Economics Are Still Challenging

Back to reality. AI requires capable data centers and vast amounts of electricity. Running AI models is expensive. And in contrast to traditional SaaS business models, where gross margins are 70–80%, AI’s gross margins are negative. That means that the more users you add to your AI service, the higher your loss. That’s because more AI usage needs more data centers and more electricity.

That means that most AI services still depend on VC money, as they are making losses. Investments into AI technology so far have been 1,000x higher than current annual revenues. If you’ve ever negotiated a valuation for a SaaS company, you should know that getting a 10x valuation on annual revenues is already extremely challenging.

The horrible unit economics of AI are the reason why everybody speaks of an AI bubble that might burst at any moment. The moment will come when the first venture capitalist pulls out of AI, and that might trigger a chain reaction. But because the technology is insanely useful and mature, it will survive a bursting bubble — just like the internet survived the dot-com crash in the early 2000s.

Agentic AI Will Be The Next Industrial Revolution

Let’s try to look ahead into the next phase of AI. I think that Agentic AI will be comparable to the Industrial Revolution. It will allow AI not just to cough up answers as text or speech, but to take actions on a user’s behalf — completely autonomously.

A fridge that orders just enough food to match the user’s individual needs and preferences. Customer service systems that don’t just regurgitate pre-recorded answers, but can understand and solve an individual user’s problem. Workflows that don’t just execute pre-recorded sequences of repeatable tasks, but decide by themselves when to execute what tasks and workflows.

Agentic AI will allow products and workflows not to be designed for audiences, but for individuals. That will eliminate tons of white-collar jobs, and it will accentuate the shift towards mass personalization.

What can you do to stay relevant if you’re not an agentic AI engineer? I’d suggest you ditch the textbooks and learn how to use a wrench.

Artificial General Intelligence Is Still Far Away

Besides Agentic AI, the other buzzword is Artificial General Intelligence (AGI). Its proponents claim that AGI will be more capable at any cognitive task than even the smartest human beings, and that AGI will be achieved before 2030.

Wait, what!?

A bunch of AI engineers, usually highly skilled in logic, mathematical models, and statistics, should create a tool that’s better at composing music than the best composers the planet has ever seen? Or the same bunch of AI engineers should create a tool that’s more empathic than the best psychologists in the world?

I’m highly sceptical that AGI will ever be achieved.

Conclusion

I might be wrong in my judgment. It’s easy to criticize somebody retrospectively for a false judgment. But it’s hard to stay flexible and adapt your judgment if events unfold differently than you planned or predicted.

With the current speed of development in the domain of AI, staying flexible and adaptable will be the key skill in 2026.