AI has been developing rapidly for several years and is constantly making new advances. Language models are getting smaller and better. Companies and tech-savvy individuals are paying more attention to what happens to their data. A forecast for developments and AI trends in 2026.
Introduction
A prediction for a one-year period was already made here for the year 2025. The forecasts were fulfilled to a rather high extent. So, the development of LLMs regarding their size and performance capabilities was well predicted. However, this is not really a surprise for all those who intensively deal with AI technology.
As predicted, agent systems are also increasingly finding their way into the everyday lives of some AI-savvy individuals. However, there is no sign of a triumphant advance: AI agents can not only be powerful, but also dangerous. At the same time, they often fail.
Autonomous systems were on the rise in 2025. However, their breakthrough is yet to come, with the exception of war scenarios (drones).
What will happen in 2026? Where will AI develop? What new things will happen?
The following is a forecast.
Forecasts for the development of AI
The individual forecasts are presented in loose order. They will be evaluated towards the end of 2026.
The rise of open source
Recent advancements in AI have been rapid, bringing new breakthroughs every year. Language models are becoming smaller and more efficient. Companies and tech-savvy individuals are increasingly concerned about what happens to their data. A forecast for developments and AI trends in 2026. Open-source AI models and methods have been improving for a while now. For example DeepSeek. DeepSeek, as a Chinese startup, even made it into the evening news on ARD. Recently, DeepSeek has brought innovative methods to improve AI models through previously unknown training methods.
Meta has decided as of December 2025 to withdraw from Open-Source. However, this shouldn't be of interest to anyone, because it will also continue without Meta. It would actually be best if Meta disappeared off the face of the earth: Its transgressions against humanity are too significant and numerous .
As before, LLMs and image models will become smaller and smaller or better. ChatGPT will not only have a problem with Google's AI dominance, but also with free AI, which everyone will soon be able to operate to some extent on their own computer system (or powerful smartphone).
Security requirements
People and companies will pay more attention to data protection and data security. Not that companies in Europe need to place particular emphasis on compliance with the GDPR: supervisory authorities are still a toothless tiger. Rather, companies are worried that their trade secrets or other sensitive data will fall into the wrong hands.
It is well known that AI systems are processing more data than ever before in human history. So it's no wonder that some people are worried. It starts with AI training to create cloud AI models such as ChatGPT or Gemini. Many billions of texts are collected from the internet for this purpose (you could also leave out the "b" here).
Three types of views on data security can be identified:
- The first group are the Acceptors and Reality Deniers. They have resigned themselves to their supposed fate: control over their own data and that in the company is not possible. But that's not so bad, because Microsoft is certainly safe anyway. Spoiler: Microsoft is not safe. But fortunately, nothing ever happens. Espionage has the task of remaining unnoticed.
- The second group are the alleged risk reducers. They think that with a ChatGPT filter or anonymizer everything will be fine. Spoiler: Such data filters do not work. This leads to an even bigger problem than before: As soon as the privacy filter is installed, some people think everything is safe now. Therefore, sensitive documents are uploaded to the chat even more recklessly.
- The third group are the Innovators and Free Spirits. They don't want to be suppressed by the US and data giants like Microsoft or Google. The solution is being sought out and utilized. Spoiler: There have been local AI as a safe and affordable alternative for several years now. The basis for this is open-source models. The application scenarios are diverse. The answer quality is often better than with ChatGPT for specific use cases.
Cyber security is becoming increasingly important. Why shoot down a missile when you can paralyze an entire country easily and almost untraceably with malware?
Thanks to Microsoft, the malware named Copilot is now frequently landing in German companies and inviting others to be exploited by villains. Fortunately, that only happens in other companies. Yours is definitely safe ([1]) !
Multimodal systems will become the standard
It all started with ChatGPT as a language model. There were also image-only models. For some time now, there have been models that combine text and images or even speech.
These models are called multimodal: They support multiple types of data.
How DeepSeek has shown with DeepSeek OCR, a machine learning model that interprets a scanned text as an image (rather than trying to extract the text directly from the image) can develop a significantly better understanding of the text. At the same time, the information content of the text can be expressed with a fraction of the tokens previously used (compression).
So in the future, we may still say language model or LLM, but mean "AI for whatever".
Enterprise AI faces the "scale or fail" moment
Many companies believe that a chatbot is a savior. It is true that chatbots are very suitable for general questions and answers. Chatbots are the wrong choice for specific applications in companies.
A specific problem requires a specific solution.
Would you ask Leonardo Da Vinci how to change a truck tire? Or would you ask a specialist?
Universal systems can only solve specific problems to a limited extent. In any case, their answers are unreliable, non-transparent and, at the end of the day, uneconomical.
Those who believe they can succeed by taking the easy route will only succeed with general problems. For example, chatbots for world knowledge, or AI assistants in software development.
Business problems will be solved correctly with preconceived AI systems. This is the somewhat more difficult path that has no shortcuts. Fortunately, this is not a problem, because thanks to AI, AI apps can be produced faster and better (and with a nicer look) than ever before.
Because data security (see above) is becoming increasingly important, and at the same time cloud AI will be labeled a failure, the decision for local AI is easy. A local AI runs on its own AI server, which is either located in your data center or can be operated as a rental server. At the same time, such AI systems can be optimized. Because the truth is also:
Many AI solutions do not use AI at all. The best AI solutions are based on AI and conventional methods.
One hype chases the next
Agent systems have already been mentioned above. They will continue to be pushed and receive increased hype. Agents are powerful. But not every problem needs an AI agent. Marketing people will still only talk about agents because nobody knows what they are. But it sounds good and maybe it sells.
A continuation of AI agents is the Model Context Protocol (MCP). This allows machines to talk to each other in a defined way and AI applications to call each other. So far, so good. But is that what we want? Anyone with a little imagination can only realize that this is not a good idea. Nevertheless, new protocols are constantly coming onto the market, such as the Agent-to-Agent Protocol (A2A) or the Agent Communication Protocol (ACP).
Computing hardware scales to new dimensions
The hardware development is keeping pace with the exponential growth of AI requirements. NVIDIA Blackwell (B100/B200) integrates 208 billion transistors on two chips, connected with an extremely high bandwidth for data exchange between the chips.
AMD concludes with the MI400 series: The chips promise 10x performance compared to the previous series, and a bandwidth increase.
The Quantum Computer-AIConvergence is gaining momentum. However, realistically speaking, there have been no significant successes in quantum computing that are relevant to our existence. It remains to be seen whether with the help of AI a new hardware can emerge that can solve the quantum problem and at the same time become another threat to humanity.
The development of AI is progressing rapidly, bringing new advancements every year. Language models are becoming smaller and more efficient. Companies and tech-savvy individuals are increasingly concerned about what happens to their data. A forecast for developments and AI trends in 2026. What will also be exciting is Edge-Computing, i.e., running AI on smartphones or other mobile devices. There are already methods that allow you to calculate the AI response directly in your browser, provided you have a graphics card capable of running AI. Smaller models even run relatively quickly on good CPUs.
AI pornography on the rise
Even Sam Altmann (OpenAI) has a plan to offer AI-generated porn in the future. A good idea from a market perspective: sex sells and you can earn a lot of money with high-quality eroticism. OpenAI needs it, given the billions lost every day from the operation of ChatGPT.
AI video generators are already delivering such good results that porn films can already be customized today. The quality will increase immeasurably. There is probably more training material available in the erotic sector than for any other genre.
Who has already tried to generate a picture with a AI-image generator according to their own own needs, knows what that means: You get exactly what your neurons want. Your script, your imagination, your result. No rummaging around for need satisfaction, but defining needs with the best possible (virtual) satisfaction. When you've seen "everything" at a certain age, it's not as interesting anymore. But for many, it will become a permanent source of fulfillment.
The consequences of technological progress
It has been clear for some time that people's expectations of AI systems are constantly growing.
AI must understand this" is something we hear more and more often.
Or take the sloppy prompts in the chat window. If the answer is not perfect, the mediocre result on social media is used as proof of the failure of the "supposed" artificial intelligence.
The job market will be turned upside down. The truth is:
- Programmers will be able to be let go in significant numbers. Namely, the poor and mediocre software developers, who can each be replaced by a handful of skilled developers who know how to use AI to achieve progress (the example shows how an otherwise barely feasible redesign was possible in short time).
- Translator will become largely irrelevant. Just use DEEPL or a language model instead.
- Web Agencies will become largely insignificant. Every tech-savvy person can create excellent-looking websites and web apps thanks to AI themselves.
- Consequently, designers will become increasingly insignificant.
- The film industry is shifting from Hollywood and Bollywood to "World". This is made possible by video generators that can deliver movie quality.
- AI will be a new employee in every company that works intensively with computers.
As in the previous year, autonomous systems will become better, cheaper and more powerful. The failure robot from Russia, which was heralded as the humanoid of the century and fell over during a public demonstration and had to be embarrassingly removed from the stage, does nothing to change this.
Warfare will be the first to "benefit" from this. Defense with the help of AI is still possible.
Outlook
Wild, uncertain and at the same time the most productive time in human history is approaching.
The negative consequences of the endless possibilities offered by AI and Automation of intelligence can no longer be halted.
The positive consequences are immense. Let's enjoy them as long as we exist.
Anyone who is not afraid of the end of human history will probably have one of the following starting points:
- No idea about AI (which is not a bad thing, at least it sometimes makes for a better life)
- Some people are tempted to think that the human race's stupidity and seemingly sometimes self-destructive urge is finally over.
- Too little imagination.
- I have no idea what "chance" means. Even professors of computer science on Markus Lanz talk down the possibilities of AI because AI is based on statistics (which is true). However, these partially underdeveloped people apparently never heard of the postulate of chance in quantum physics. Quote from the source: "Special significance for the development of philosophical conceptions of chance had the dialectical-materialistic interpretation of the results of quantum mechanics, which proved the objective existence of chance for physics." See also radioactive decay. Probabilities are the precondition for intelligence, not the reason why no intelligence is possible.
- Parents of children: Understandably, (most) parents want their children to have a good life.
The latter group in particular will probably be disappointed if they live longer than 10 or 20 years. Because by then, the problem with pensions will have been solved.
Until then, we should take advantage of the opportunities offered by AI systems. Automation of business processes will become an economic factor that will provide a boost to growth.
My name is Klaus Meffert. I have a doctorate in computer science and have been working professionally and practically with information technology for over 30 years. I also work as an expert in IT & data protection. I achieve my results by looking at technology and law. This seems absolutely essential to me when it comes to digital data protection. My company, IT Logic GmbH, also offers consulting and development of optimized and secure AI solutions.
