2025 AI Predictions: Agents, Opportunities, and Threats

We’re living in remarkable times. By 2024, AI has already transformed many aspects of business and daily life. This momentum won’t slow down in 2025—in fact, it’s set to accelerate. Since 2,356,845 AI experts have already shared their predictions for 2025, I won’t sit on the sidelines, and share my thoughts as well.

So, what are the key points to focus on speaking about the directions of AI development as we step into 2025?

But first, let’s briefly recap what has happened in the world of AI in 2024

Not everything by any means. Just a few key points I want to focus on in this article.

Multimodality

LLMs became much more capable in terms of multimodality, input and collaboration options. This opened up multiple new opportunities and use cases.

LLM improvement slowdown

LLMs became more capable. But their progress slowed down. Developers shifted their focus toward creating smaller, faster, cheaper, yet still more capable models. The priority moved from building the best model on the market to creating models that could be widely adopted by businesses and individuals.  

Content creation industry landscape has changed

Generative AI tools dramatically transformed content creation in 2024. AI images have already almost got rid of the majority of the essential issues. AI videos—not yet. AI audio showed impressive progress while also triggering a copyright nightmare.

OpenAI was confidently losing its leadership

OpenAI obviously stopped being an undisputed leader in the world of AI. At the end of 2024, maybe not even a leader at all. Updates firing duel in December 2024 between OpenAI and Google seems to be obviously lost by AI pioneers.

But Google was not the only one eating the past OpenAI advantage. Anthropic, Amazon, NVIDIA, Meta, and some niche competitors made impressive strides and laid the groundwork for further success in the AI battlefield.

AI agents became more capable

However, there are still significant limitations to achieving breakthrough progress. Companies like Google and Microsoft began integrating AI agents into their ecosystems. CRM developers introduced some. However, it still feels like an early stage of development.

SEO is not dead. Yet.

Chatbots had some impact on SEO, but nowhere near close to killing it. OpenAI also introduced its Search feature. Still in the early days, and haven’t really disrupted the landscape yet.

Google and Microsoft Bing search engines have also introduced AI-generated search snippets. But at the end of the day, the effect wasn’t as massive as some expected. Such snippets are mostly actively displayed in a limited range of cases and primarily in a limited list of industries. And not to forget, generating such snippets costs much more to search engine owners than regular snippets. 

Some SEO tools, like Ahrefs and Semrush, have also introduced tools with AI Overviews data for SEO analysis. But for obvious reasons, the accuracy of such data is questionable. And just recently Google introduced an advanced blocking system for AI Overview checkers. Thus, I wouldn’t recommend making strategic decisions based on this data. It can be highly inaccurate.

AI hasn't enslaved humanity. Yet.

Some AI models have already shown their temperament. Now we know that they are ready to lie, just not to be turned off. So maybe at some point, we’ll encounter some incidents. But 2024 has ended, and humanity is still not enslaved by AI. At least not directly.

2025 AI Predictions

If you’re naturally full of ideas that you struggled to bring to life before, this year might turn out to be much more exciting than you think.

So, here is what we can expect in the AI world in 2025, in my humble opinion.

Rise of AI agents

Many expect 2025 to be the year we see a massive rise in the use of AI agents at all levels. 

But first, let’s clarify what qualifies as an AI agent. Simply put, it has three key characteristics: 

• It uses AI.
• It can perform multi-step actions to achieve goals by using various tools.
• It has a memory.

So, basically, today we already have all of the above. And many AI agents already exist. We also have AI agents building platforms, like CrewAI, that make it significantly easier to build an agent on your own without deep technical expertise. 

But why haven't we seen the exponential rise of AI agents yet? Like the growth of interactions with AI agents to somewhere near the number of interactions with chatbots. There are two reasons for this:

• Still, there are some difficulties in connecting various tools. This experience is not 100% seamless yet. But it’s progressing in the right direction;
• There needs to be a genuinely high demand for AI agents both from businesses and individuals. And an efficient business strategy behind this. 

Is there potential for exponential growth as early as 2025? 

I think it's possible. Closer to the end of the year. Or in 2026. 

Just before publishing this article, OpenAI introduced its Operator and Agents. First, it’s available only to Pro users ($200/month) in the USA. Basically, for testing purposes, before launching it to a broader audience. However, this preview provides some insight into OpenAI’s vision for agents and its approach to managing related privacy and security concerns.

The number of platforms for creating AI agents is growing. Agents are quickly becoming more mature. Big players, like Google (with Vertex AI), are already in the game. Integration options are expanding.  

AI agents in CRM

Key players in the CRM market are also actively digging into developing AI agents. 

In September 2024, HubSpot announced its version of AI agents and shared its vision. Salesforce launched Agentforce.    

Still in their early stages. Need to monitor developments and updates. By the end of 2025, customer relationship management will inevitably undergo significant changes compared to how we remember it today.

What blockers for AI agents will stay in 2025?

One of the obvious ones is the completion of actions across multiple tools. Full control is something that is still difficult to achieve due to required permissions and security considerations. It feels like we’re currently 1-2 years away from fully functional agents.   

Why should businesses use AI agents?

That’s quite obvious. AI agents allow you to do more than your competitors (who aren’t using them) within the same amount of time. And usually do the job better.

Great. Now we are faster. AI performs numerous tasks. Can we now fire humans to reduce expenses? Well, you could. But that sounds like a terrible idea. Your human team is now free from routine tasks that can be automated. They can focus on what truly matters. So why not keep humans and expand their role in addressing more meaningful challenges, ensuring your competitors don’t close the gap you’ve already created?   

Many businesses are already aware of this. If you haven’t yet, now is the time to catch up.

Smaller, faster, and cheaper LLMs should help with the efficient business model behind developing and selling AI agents.    

Active promotion of the AI agents market by the leaders of the AI industry and thought leadership should also make an impact.

However, one bottleneck will remain: security concerns, which may prevent some platforms from opening their tools to everyone. For instance, will Booking.com grant access to its hotel booking system for all AI agents, given the potential actions of bad actors? We’ll see. 

Challenges in selling AI agents

Be ready to face specific challenges selling your AI agents. The issue is that the people most qualified to evaluate whether your agent is good and worth purchasing are often the same people these AI agents will replace.   

For example, if you’ve developed an agent that performs initial screening of job applications, the business owner may not be the best expert to evaluate your product. The HR specialist might be that person. And this is the very person your AI agent would replace. Imagine what they might tell their boss about your product.

Will 2025 be the best or worst time to start a SaaS project?

It has never been easier to start your own SaaS product, with the help of AI. So, jump into this right now without hesitations?

Well, better think twice. Satya Nadella, the CEO of Microsoft, probably wouldn’t advise you to do so.

Consider what he has to say on this topic:

Shortly - AI agents will be used to interact with data and deliver it to you in a format pre-programmed by you (or the agent’s developer). This means that the full SaaS experience, with its user interface, may no longer be necessary. The only truly valuable component of a SaaS offering will be a properly structured dataset, prepped for interaction through an AI agent.

So, better think about this. This doesn’t necessarily mean all existing SaaS products will vanish soon, or that new SaaS ideas will become obsolete. Just think once more about your case.

Best time to bring your brave business ideas to live

The implementation of the boldest ideas in record time will become more feasible than ever before. The lack of deep technical skills will no longer be a blocker. Ability to start and scale business successfully with a very small team or even solely - a critical game-changer if you start with a tiny budget. 

Remember, you’ll have an option to hire (and fire) AI agents. 

Will we witness the rise of the first one-person unicorn company?

Yes - 51,36%. No - 48,64%. 

Do you intend to become the very first one?

Primary directions for LLM development

Multimodality

Multi-modal LLM capabilities will change AI usage much more dramatically in 2025. Multiple options for input and higher-quality data processing open numerous exciting opportunities for businesses and consumers. 

Interacting with AI tools will keep becoming more natural and seamless. Here it’s even hard to predict where it will take us by the end of the year.

Fine-tuning for specific niches

Many new LLMs will be specifically fine-tuned for narrower niches. This trend existed in 2024. But a list of LLMs released at the end of 2024 is definitely a sign that 2025 can be a real breakthrough in this aspect. Just check the list of models released by Amazon in December 2024, and imagine what’s coming.

Focus on cost-efficiency, not maximum capabilities

At the beginning of 2024, it still felt very much like a competition to release the best capable LLM on the market to prove your leadership status. However, by the end of 2024, it became clear that the focus had begun to shift.  

The best recent models are now capable of solving enormous reasoning tasks. But completing such complex tasks is also very expensive. I mean really expensive. Way more expensive and energy-consuming than generating videos in Sora in the $200/month package. Focus on building a model affordable for only a handful of users is not a sustainable business strategy.

What’s the point of completing a task with AI if doing so could black out a small suburb of Los Angeles? To reach Artificial Super Intelligence (ASI) first? Yes, it is an ambitious goal. Will the one who achieves it first take control of the entire world? Yes, it is quite likely. 

But the closer an LLM gets to Artificial Super Intelligence, the more expensive each next step becomes. Accordingly, the issue of funding becomes increasingly critical. Therefore, to achieve ASI, it is essential to have a viable business model for the mass adoption of LLMs.

By the end of 2024, we saw a clear move toward developing smaller, faster, cheaper, yet significantly more capable models than shortly before. At the time of writing, Gemini 2.0 Flash (Experimental) is probably the best example. 

This shift paves the way for the broader application of LLMs, making advanced AI solutions accessible to almost everyone. And I think this trend won’t go anywhere. In relation to building AI apps and agents in 2025, be ready to hear the word “latency” more and more frequently. 

Open-source LLMs 

I have always seen potential in open-source models. But it wasn’t obvious if the most capable of them would be actively used by a broad audience. Because of the main cornerstone - hardware resources required for running such large models locally. 

But luckily, closer to the end of 2024, open-source LLM developers started actively releasing smaller yet capable models. Still a bit too big to run them on your average laptop frequently. 

Meta, with its family of Llama models, will probably remain the main player. But it’s not certain. Especially after the release of DeepSeek-R1 on January 20. NVIDIA has already released a couple of interesting models and definitely has the most affordable resources for training models among all players on the market. 

Also, I consider the Mistral team one of the best LLM developers overall. Fewer resources, less hype. But at certain stages, in some aspects, it was better than competitors. Initially was built based on smarter principles. And I have always liked software products developed by French and Dutch teams. Usually robust and reliable. Unfortunately, perfectionism takes time. And in the AI world that’s the resource you're always short of.

Proprietary vs Open-source LLMs

While I’m optimistic about open-source models, I think in 2025 proprietary LLMs will still be more popular. However, this year will be the last year of their dominance. Even their current dominance was called into question with the release of the DeepSeek-R1 LLM, which has already caused hysteria among the big players. Followed by the release of Janus-Pro-7B on January 27. Anyway, by the end of 2025, we may witness a significant shift.

Key points to consider

Open-source LLMs (like LLAMA, DeepSeek, Mistral, etc.) are totally competitive with proprietary models, like GPT, Gemini, etc.

The use of LLMs is not cheap. When the quality of output is roughly the same, cost-efficiency and latency become the deciding factors.

Using open-source LLMs locally is significantly, significantly cheaper than proprietary ones.

Since the release of the first open-source models, the main obstacle to their widespread adoption has been the hardware requirements needed to run these models locally—at home or in the office. For most people, this hardware was unaffordable. As a result, “renting” solutions in various forms was the most common alternative.

By early 2025, we can already say that the bottleneck related to computational power for open-source LLMs will soon be resolved. There are strong reasons to make this claim.

In January 2025, one of the most notable events at CES 2025 in Las Vegas was the NVIDIA Keynote. Among the many important announcements and innovations presented by NVIDIA CEO Jensen Huang was a supercomputer specifically designed to work with AI and, in particular, with local LLMs. Its provisional name is NVIDIA Project DIGITS.

This computer was created specifically for consumer use. It can be used as a home server, workstation, or personal cloud where your local open-source LLM will reside and handle all your home/office tasks. It’s not only designed to run LLMs but also to train them! 

Such an AI supercomputer will be able to handle LLMs with 200 billion parameters effortlessly. 400 billion parameters? Just connect two devices! Moreover, the entire NVIDIA software stack is there out of the box. And perhaps most surprisingly, this computer is roughly the size of an 8TB external HDD.

The cost of this game-changing device will not be astronomical, starting at $3,000. It is expected that NVIDIA’s AI supercomputer will be available for purchase as early as May 2025. While it will still be out of reach for some, it will also be affordable to enough people to make it clear that the future belongs to local open-source LLMs.

Here’s a segment from the NVIDIA Keynote showcasing the new product:

Overall, I would recommend dedicating time to watching the entire NVIDIA Keynote. It will give you a better understanding of the overall direction AI is heading. NVIDIA is one of the key drivers of progress in the AI space. Not only do they quickly adapt to major trends, but they also set them.  

Among other announcements, Jensen Huang revealed the release of three LLMs: Nano, Super, and Ultra. All of them were trained using open-source Meta models—LLAMA Nemotron. It’s not hard to see that one of the major players has already placed a bet on open-source LLMs.  

The race for superiority

In 2025 big tech companies will keep investing a lot in their AI tools without a 100% clear monetization strategy. Just to capture the largest possible share of the market in key AI niches and establish oneself as a market leader, setting AI trends and standards. Even though currently they are spending much more when getting back. The winner’s reward can be immense. 

Nobody is ready to stop. 

Somebody, like Apple, has not even really started. Despite silently acquiring multiple AI businesses.

Will 2025 be the year of OpenAI’s sunset?

I think OpenAI will obviously lose its leadership and flagship status in the AI world in 2025 unless something really unexpected happens. 

Just 2 years after releasing the product that has changed the world, OpenAI seems unable to compete with bigger companies, even being backed by Microsoft financially and hardwarely. 

Key personalities who built this company from the ground keep leaving it on a regular basis or plan to do this for various reasons:

• OpenAI co-founder Ilya Sutskever and safety leader Jan Leike announced their departures in May 2024.
• In August co-founder John Schulman decided to leave OpenAI and join its direct competitor - Anthropic.
• In September, Chief Technology Officer Mira Murati announced her departure. Bob McGrew, the Research chief, and Barret Zoph, a research vice president did the same.

And these are just major announcements.

Consequences are already there.

12 days of announcements

OpenAI’s pre-Christmas 12-day announcement marathon was supposed to be massive. But was it at the end of the day? Let’s recap the new products/features they unveiled.

1. New o1 model

Ok. It was announced as a better one, compared to the previous one - GPT-4o. Is it obviously better than competitors’ flagship models? Seems like not. Has a faster mini version, which is good. Basically, a regular LLM update in 2024.

2. New o3 model

Still not publicly available. But announced as capable of solving highly complex reasoning tasks, which takes a lot of time and costs a lot. Any practical usage for a broad audience? 

3. Sora

First announced in February 2024, Sora finally became available for paid ChatGPT users in December. 

Expectations in February 2024: Wow! This is a complete game-changer. High-quality videos 1-minute long. A Hollywood killer. Now anyone can make a movie. Other generative video tools like Runway or Pika will simply die.

Reality in December 2024: Up to 20 seconds (in a $200 plan), still hallucinates, lack of controls. It seems other generative video tools used these 9 months far more effectively. 

And one more negative consequence. Sora was not (and at the time of writing, still isn’t) available even for paid users in Europe. This angered many European OpenAI users and triggered some to cancel their ChatGPT subscriptions, switching to competitor products, like Gemini and Claude. But of course, nothing will stop them from using ChatGPT for free.

4. Canvas    

Canvas update was a promising release. A couple of interesting new features. Multiple use cases. If they truly have more ambitious plans developing this tool, it may become really helpful and popular.

5. Advanced voice

One more promising feature in case of further development and integrations.

6. ChatGPT search

The Search feature is also not new. Currently, just an enhancement of the Chatbot. Will it evolve into something bigger? Not sure. ChatGPT didn’t become a Google killer. And it’s unlikely to compete with Google in the traditional search engine market. And most likely the main OpenAI investor, Microsoft, would not like the idea of having one more competitor for Bing.  

7. Where OpenAI is still able to shine

Despite all my pessimistic thoughts about OpenAI 12-day announcements, I must admit that there was one release that completely blew my mind. OpenAI has pioneered one more feature that it can be truly proud of. You can now use ChatGPT with a rotary dial phone. Haven’t had one in 20 years? Doesn’t matter—just visit an antique shop and grab one. 

But if OpenAI doesn’t find a good business model to monetize this feature, they have another area to focus on—creating a humanoid robot. The right time, the right place. I can easily imagine such a robot answering incoming calls through a rotary phone.

Operator, Agents, and Scheduled tasks

On January 23, OpenAI introduced its Operator and Agents for limited public testing, available to Pro users. And shortly before that, it released the Tasks feature in ChatGPT. Quite promising tools for building more advanced agents. But of course in the very early stages. Will see how these developments evolve within OpenAI’s ecosystem. 

More problems for OpenAI

OpenAI is trying to shift from a non-profit to a for-profit status to overcome challenges and push its growth forward. But some obstacles could prevent this from happening.

OpenAI early investor Elon Musk is trying to block this potential status change in court. And, considering Musk’s relationship with Donald Trump, the chances of an unfavorable decision for OpenAI seem quite high. 

Microsoft may also have its own interests and considerations that do not necessarily align with Sam Altman’s plans. Some key employees are leaving or planning to leave, as they prefer to work on a non-profit project with a focus on safety.

Do bigger tech companies, like Google, Meta, or Amazon, have such blockers? No. Do they have more resources? Yes. Are they already moving faster? Seems so.

A significant threat to OpenAI has recently emerged from China. The true backers of the company DeepSeek remain unknown. However, when it comes to China, it’s not hard to guess who the ultimate beneficiary is and who provides comprehensive support.

DeepSeek's models are roughly as capable. They reach the level of OpenAI’s models with a slight delay. However, they are much more affordable. The reason is the use of less advanced but significantly cheaper chips (as declared, but not necessarily true), some smart solutions, and, of course, the labor costs.

So should we expect OpenAI to sunset already in 2025? 

I’d say yes. It’s unlikely that this process will be fully completed in 2025 since OpenAI still can turn the game in their favour focusing on humanoid robots and rotary phones. Or perhaps they’ll introduce something groundbreaking that shifts the narrative entirely.

The number of low-quality digital products will rise

Creating various AI-based solutions and products using such solutions is becoming more accessible every day. Their number is growing exponentially. Today it’s already quite difficult to choose the robust ones. By the end of 2025, this problem will become a complete headache.

Hybrid employees (Human + AI) 

Many companies, starting with large and more tech-savvy, will begin implementing hybrid workflows and workplaces (Human + AI) with increasing involvement of AI. 

Once again, this doesn’t necessarily mean that AI will replace humans. When used correctly, it will: 

• Take over routine tasks that can be fully automated.
• Act as a human assistant, working very closely to make people incredibly productive.

So don’t be afraid that AI will take your job. Learn how to work with AI in pairs, in small groups, in “Swedish Families”, or as part of larger teams. And rest assured, in 2025, only humans will be enjoying vacations, paid sick days, and other traditional human perks.  

AI onboarding

You’ll start hearing this term more and more frequently. AI onboarding will involve quickly training newly "hired" AI agents using your internal documentation and datasets.   

Copyright nightmare

Many warned about copyright issues right after the first AI-generated images were published.

Now, AI can also generate videos, music, and much more. And AI tools gained these abilities (and continue to improve) by training on real images, videos, audio, music, etc., created by real people who legally hold copyrights (at least in theory). However, they are unable to effectively protect their rights for obvious reasons.  

Two years into the AI era, and we are already facing total chaos and a nightmare for artists and content creators regarding copyrights and everything related. 

Here let’s just research the case of the impact of AI on the music industry. To understand the issue, I recommend watching this video:

Briefly about the key points in the context of Youtube:

• How does YouTube determine that you have the rights to specific music? You were the first to upload it to Content ID.  
• AI tools allow users to generate music in the style of a specific artist by using their music as a reference sample.  
• The generated result will sound like the music of that artist.  
• Users can upload this result to Content ID.  
• When the artist uploads their new music to YouTube (whether as a song or background music), it may be recognized as a copyright violation against the user who generated music in the artist’s style using AI. 
• This has already become a lucrative business for bad actors.  
• On YouTube, artists can do very little to protect themselves in such situations.  
• YouTube is a private company. Its system is automated, and they don’t want to deal with it manually, even for a channel with 500,000 subscribers.

Is there a way for artists to protect their rights in the age of AI? 

It depends on the country and its laws.  

In the U.S., for example, licensing music before uploading it to the internet can help. But this won’t completely eliminate potential issues on YouTube.

What impact will AI have on the creative industry in terms of Copyright?

Negative. Until this is clearly regulated by governments.

Will artists stop creating new music?

Some might. And that’s sad.

Will artists stop releasing digital content and focus on live performances? 

We’ll see. As a form of protest—it’s very likely. 

Will Woodstock come back as a result? 

I hope so.

AI in healthcare - advanced LLM will make a breakthrough 

The year 2024 demonstrated how useful AI can be in various aspects of healthcare. This is one of the few fields where the use of the most capable and extremely expensive LLM, like o3-Pro, makes sense.  

I expect that by the end of 2025, we will hear about several breakthroughs in healthcare. Not some minor improvements, but significant breakthroughs. Those that humans couldn’t achieve without AI but can now accomplish with its help. 

AI Wearables will start actively coming to the market

Not an area of AI I follow very closely. The current solutions are still quite underdeveloped. And not widely spread. That said, big players like Apple and Meta are in the game. So, I’d expect to see real progress at least with smart glasses by the end of 2025.

Will SEO die in 2025 because of AI? 

Of course. Why not? SEO is dying very frequently. It has a habit of “dying” after every minor change in the industry. Over the past 17 years, I’ve seen many "deaths" of SEO. The last time it "died" was after the release of ChatGPT in 2022.  

Seriously. No. SEO won’t die in 2025. 

However, the most productive SEO specialists will need to adapt their approaches, learn how LLMs work, and adjust content, SEO, and digital marketing strategies to align with the rapidly changing AI landscape. 

And for those who still haven’t accepted this fact yet: chatbots are also search engines. Just utilize them as search engines.

What will happen to AI snippets in SERP?

To understand the future, you first need to understand the nature of AI Overviews.

Right after ChatGPT’s release, it was called a "Google killer," even though it wasn’t yet connected to the internet.

Google didn’t have an immediate prepared response. So the best idea they came up with was to introduce AI-generated snippets to SERPs, later referred to as AI Overviews.

Did people switch to using AI Overviews in Google Search or their equivalents in Bing en masse? Not quite. The hype around AI in search engines has faded. Chatbots are now connected to the internet, and conversational searches can be conducted in a more convenient environment. 

AI Overviews still exist but are actively displayed only for a limited range of topics and certain user intents.  

Generating AI snippets in SERPs is significantly more expensive for search engines than regular snippets. But it doesn’t necessarily provide greater value. So it’s a weak business solution. 

In 2025 the number of AI Overviews will drop significantly

As of early 2025, it’s becoming clear that AI-generated snippets have served their purpose—to show that "we’re good at AI too." There aren’t many reasons to keep them in their current form.   

So they will only be used where this format best aligns with user intent and in niches where search engines can safely take responsibility for generating a full response. Being an "answer machine" is what Google has long wanted to become.  

However, AI will play a significant role in SERPs by generating various types of enhanced snippets, such as shopping cards. And most likely, we’ll see plenty of new experiments in this area.

AI impact on link-building

In 2025, the influence of links on website performance in traditional search engines is unlikely to change dramatically. However, what has already changed and will continue to evolve rapidly – is the ability of search engines to identify unnatural links using AI.

So, in 2025, when selecting link donors, I would advise focusing less on domain or page authority metrics and more on traffic—both incoming and potentially outgoing traffic (to your site). Otherwise, the chances of wasting your link-building budget will increase sharply.  

As a consequence of the above: Goodbye, all the mom's link builders spamming mailboxes with offers of “high-quality” backlinks.

How will AI change Web Development?

Will AI be able to build an entire website just with one prompt or a sequence of prompts in 2025?

Well. It has been able to do this since 2023. And at the very beginning of 2025, it can do it in multiple different ways, using various tools. The problem is that it’s still a kind of Lego for kids. Building more complex projects still comes with many concerns, starting from security. Such websites also fall short in scalability, maintainability, SEO, and other critical aspects.  

What usable website will you be able to build with AI by the end of 2025?

Within a one-year timeframe, I can see more or less usable small websites built with AI in the environment of hosting companies and some platforms for kids, like Wix. In the format of “build and forget”. Just if you need a couple of pages with a good design and have zero SEO ambitions.  

The shift from traditional to headless CMS

This is one of the possible trends for 2025. With the process of creating applications becoming more accessible to less technical specialists, more businesses (and individuals) will have the opportunity to interact with their target audiences in different ways. For example, in addition to your website, you could create a mobile application. 

That’s where headless CMS will become handy. Content, as a single source of truth, will live in the backend (“body”). And you can serve different front-ends (“heads”), such as websites and mobile apps.

The development of many custom niche products will make little sense

An example with the hospitality industry product - Amenitiz. It offers a website builder, booking engine, property management, price advisor, and some other features. All these are perfect candidates to be replaced by AI agents integrated with open-source CMS platforms.    

Speed-up website content editing with voice and smart glasses  

Tired of adding content to your website CMS manually? Maybe good times are just around the corner.

Multimodal AI capabilities are getting better and better very fast. So robust solutions of adding content in your CMS using voice look quite achievable in 2025. 

But can we move even further? Smart glasses packed with AI features already look quite promising. And some are even affordable. 

So imagine using smart glasses + voice. With glasses, AI can see what you’re looking at - your screen with a CMS dashboard. And you’re just dictating the instructions, making edits or adjustments in a conversational format. 

The productivity of every employee equipped with such capabilities skyrockets. The need for content managers performing monkey jobs decreases. A marketer can add content on their own without spending significant time on this while ensuring everything looks its best.

Legal limitations for AI are coming

The AI era arrived quickly, and governments were unprepared. Now they are trying to catch up by developing regulatory directives for AI. However:  

• They are lagging, often playing catch-up.
• They have a vague understanding of how to regulate AI properly.  

But the era of AGI (Artificial General Intelligence) is drawing closer. If not in 2025, then likely in 2026. And ASI (Artificial Super Intelligence) is on the horizon as well. 2025 will likely bring certain preemptive regulatory restrictions. Undoubtedly, some of these will become obstacles to AI development itself.  

What will be most interesting is to compare regulatory approaches across key regions. In the European Union, as usual, strict frameworks have been developed, similar to earlier regulations for cryptocurrency.  

In contrast, U.S. President Donald Trump has declared a regime of maximum support for the development of AI and cryptocurrencies. He even launched his own cryptocurrency, the $TRUMP meme coin, along with his wife’s meme coin, $Melania. However, completely unsubstantiated, which shocked even the representatives of the crypto industry who supported Trump.

But, as with all of Trump’s other statements, it’s worth waiting to see what actually materializes and what remains in the annals of history as empty campaign rhetoric.

Irreversible consequences for humans

AI will make intelligent individuals with a continuous drive for self-education and a problem-solving mindset nearly almighty. 

At the same time, for lazy people, AI will become merely a tool to minimize their efforts in learning and problem-solving. This will lead to their accelerated degradation, making them even more dependent on AI and helpless without it.  

Who will lose their jobs because of AI in 2025?

In 2024 many employees have already lost their jobs because of AI. At the beginning of 2025, it’s obvious that AI tools can already do some jobs faster and better than humans. And it cost less. Don’t demand lunch breaks, paid vacations, gym memberships, or much else. Rest assured, in 2025, AI will become even more capable and replace even more humans.

If your job can be easily performed by AI without requiring strict human oversight, you are at risk. The list of such tasks will continue to grow exponentially. Here are a few roles that will be at risk in 2025:  

1. Office clerks performing routine tasks. Especially digital ones.

2. Data processing employees. It’s obvious that in this field AI is already demonstrably better.

3. Customer support specialists.

While I believe the best approach in 2025 will be to use AI and humans in connection to provide the best customer support, it’s easy to predict that many companies will just try to to replace humans with AI tools as much as possible.

4. Junior copywriters, translators, proofreaders, graphic designers.

5. Junior developers, software engineers. 

Google recently stated that 25% of their new code is generated by AI, then reviewed and approved by engineers. And I would bet that this share will only grow. This basically means that only mid-level and senior developers capable of reviewing complex code will remain in demand. However, mid-level developers must put in significant effort to advance to senior levels ASAP.  

6. Truck drivers, delivery drivers, couriers. 

They will be partially replaced by drones and autonomous vehicles. For now, only in countries where the legal framework is favorable. Primarily in the U.S. This is unlikely to happen on a large scale in 2025, but the trend will become more apparent.  

7. Presidents, prime ministers, etc. 

Can be easily replaced by talking digital avatars with digital signatures. But, yeah, agree - dreams, dreams.

So what to do if you’re among the listed ones?

Prepare to send out resumes for hundreds of job openings and pray that they catch the attention of an almighty (and maybe soulless :) ) HR specialist on the other end?

Well, that may not be the case anymore. Many HR specialists without really valuable skills will soon join you in this CV-sending adventure. Because they will also be among the first to be replaced by AI tools. 

In fact, applicant tracking systems (ATS) existed long before the AI era. But with AI support, the process of evaluating resumes can be done faster and more effectively (for example, machines won’t subjectively reject candidates). And with the help of AI agents, the process can extend far beyond simple tasks.    

So your job is to understand which roles will be in demand in the AI era. Gain initial expertise in that field. Most likely, AI would be able to help you with this. Learn how to apply for jobs in the new conditions to get selected by AI tools.   

Or wait for the introduction of Universal Basic Income (UBI).

Will AI enslave humanity in 2025?

No, AI won’t enslave humanity in 2025. Not yet. But this prediction is not certain.

How will AI change the world in the long-term perspective

Personally, I prefer to use an agile approach and avoid preemptively modeling a distant future.

But if you like some radical forecasts on how dramatically AI will change the world, feel free to explore the article “Capital, AGI, and human ambition”. In some aspects, it might seem frightening or unrealistic, but much of it makes sense.

Or, if you’re too lazy to read, you can watch a video with commentary by Matthew Berman:

But if you’re also too lazy to watch videos…

Here are the key points:

1. Shift from labor to machines

AGI will replace humans in performing the majority of tasks. Human labor will no longer be essential and will become of little value. 

2. The power of capital ownership

As AI systems take over jobs, owning capital, such as AI infrastructure, will become more critical. Controlling these assets is the key to unprecedented economic power. If you are rich at the point of no return, caused by achieved AI capabilities, you'll become even wealthier.

3. Challenges without capital

AI systems will dominate work and decision-making. People without access to capital may lose their influence. And struggle to achieve essential results. Those who are not wealthy at the point of no return will no longer be able to become rich.

4. Consolidation of power

Owning AI-related assets could lead to concentrated power, strengthening existing hierarchies and limiting social mobility.

5. Lower focus on human welfare

AI will handle many essential tasks. Human labor will become less important economically. Organizations may pay less attention to human welfare as a result.  

6. Barriers to wealth redistribution

Efforts to balance wealth and capital ownership may face severe resistance. Especially from those benefiting most from AI advancements.

7. Universal Basic Income (UBI) introduction

Replacing humans with machines will cause severe unemployment and economic disparities. To solve this, governments might introduce Universal Basic Income. By distribution of AI-generated wealth to support the populace.

8. Threats to creativity and agency

The leveled role of human work could oppress creativity, ambition, and achieving personal goals.

9. The need for action

It's critical to avoid the listed risks. Need to develop policies that promote fair AI benefits and keep society vibrant and innovative.

But let’s not overstate the gloom

It is human nature to disbelieve negative scenarios and focus on the idea that everything will turn out well. Therefore, I do not doubt that most people will be enthusiastic about the prospect of receiving Universal Basic Income (UBI) and the possibility of never working again.  

Let’s try to imagine the future if this prediction comes true. So, the year is 2029. Headlines from online publications in different countries. Written by AI, of course.

Spain

“Spain has issued an international arrest warrant for a French citizen who attempted to book a hotel in Barcelona for tourism purposes”  

Reminder: after years of fighting overtourism, Spaniards now enjoy UBI so much that they don’t need tourist money. So, foreign tourism has been made illegal to protect their quality of life. Only virtual tours using augmented reality are allowed.

Germany

“An organized crime group in Germany was arrested for planning to organize an underground workshop for making birdhouses”  

Reminder: human labor without a license issued by the AI Committee for Labor of Biologically Passive Life Forms has been prohibited in Germany since January 1, 2028. 

Belgium

“Eddie Mercx Returns from Retirement—and Clinches a Surprise Victory with ‘Human-Powered’ Pedaling”

In a shocking twist at the Paris-Rubaix, cycling legend Eddie Mercx made a grand comeback and stunned the field by crossing the finish line first using nothing but his own leg power. His nostalgic, manual pedaling—which defied the “Automated Cyclist Uniformity Directive”—was hailed by fans as a bold return to classic cycling. However, officials quickly intervened, classifying his feat as “unauthorized kinetic authenticity.” Despite his win, Mercx was promptly detained and faced disqualification, along with a mandatory course in digital pedaling techniques.

Montenegro

“In Montenegro, a man was arrested for attempting to get up from his couch, with a 0.82% probability that he will take actions that could be interpreted as an attempt to work”  

Reminder: Montenegro was the first country in the world to take action in light of the imminent introduction of Universal Basic Income (UBI). On March 1, 2025, an amendment to the Constitution prohibited everyone from working starting July 1, 2025. The explanatory note for this law briefly stated: "Why bother?!" Since there was little time left before the ban took effect, all new laws and penalties for their violations were passed in haste and were made as simple as possible.

United States

“Man Arrested for Attempting to Sell Homemade Burgers on 4th of July”

In a shocking turn of events, Mr. John Smith was detained by the AI-Labor Bureau after being caught flipping burger patties in his backyard without a valid “Non-Robot Culinary Permit.” The Department of Patriotically Automated Celebrations reiterated that “All grilling must be conducted solely by licensed AI in accordance with the Fourth of July Non-Human Barbecue Act of 2028.”

United Kingdom

“Woman Fined for Trying to Brew Her Own Tea”

The Royal AI Council warned citizens that the practice of making tea by hand violates the “Preservation of Automated Tradition Act.” Local authorities arrested the suspect for “historic beverage tampering,” reminding the public that all tea preparation must now be managed by authorized robotic teapots.

Japan

“Unlicensed Origami Craze Leads to Mass Detention”

Authorities in Tokyo cracked down on an underground group folding paper cranes by hand. Under the “AI Artisanship Ordinance,” all traditional crafts must be performed by government-approved origami robots. The group was charged with “willful human craftsmanship” and faced immediate reeducation in a Virtual Origami Lab.

Russia

“Hobby Distillery Raided: Grandfather’s Vodka Recipes Seized”

In Moscow, a 90-year-old man was arrested for using a handwritten family vodka recipe. Officials stated that all distillation must be supervised by AI to ensure maximum efficiency and “enhanced national productivity.” The man’s neighbors reported him after noticing an abnormal lack of humming from his AI-distiller at night.

Italy

“Rome Outlaws Handmade Pasta Parties as ‘Artisanal Rebellion’”

Three Italian grandmothers were taken into custody after hosting a secret “Pasta Night,” using non-robot kneaded dough. Under the “Automated Culinary Protection Act,” humans risk imprisonment for ignoring AI pasta-machine protocols. Authorities seized rolling pins as “evidence of willful gastronomic defiance.”

France

“Parisian Cafe Shutdown for Serving Coffee with Human Touch”

A beloved corner café faced closure after inspectors discovered a barista hand-pouring espresso. The French Ministry of Automated Gastronomy declared, “AI must manage all national coffee operations to preserve cultural integrity,” adding that human-perfected latte art is “no longer authorized in the era of universal equality.”

Canada

“RCMP Arrests Man for Excessive Apologies in Unlicensed Therapy Session”

A Calgary resident was accused of “attempting emotional labor” after he was caught apologizing repeatedly to cheer up a friend. Local authorities reminded citizens that “emotional support must be dispensed by AI-registered e-Therapists,” and that unsolicited human empathy carries a hefty fine.

Brazil

“Samba Dancer Detained for Practicing Steps Without AI Clearance”

In Rio de Janeiro, a spirited dancer was arrested mid-performance during an unsanctioned Carnival rehearsal. Under the “Robotic Rhythms Initiative,” all dance training must be conducted by licensed AI choreographers to maintain “precision and universal fairness in public festivities.”

Sweden

“Stockholm Family Charged with Hand-Assembling Furniture”

In a covert operation, an entire family was apprehended for manually constructing a new bookshelf from unapproved flat-pack pieces. The Swedish Automated Lifestyle Agency noted that “only authorized assembly-bots” can legally put together furniture, emphasizing that human labor undermines the “Universal Flatpack Equality Agreement.”

Austria

“Waltz Instructor Apprehended for Leading an Unlicensed Dance Class”

Reminder: Under the “Automated Culture and Choreography Act” effective since 2027, all dance instruction in Austria must be conducted by state-certified AI waltz-bots. Unauthorized human-led waltz movements are now considered a breach of national rhythmic harmony.

Switzerland

“Citizen Detained In Zurich for Manually Adjusting His Watch”

Reminder: In Switzerland, any attempt to set a mechanical watch by hand violates the “Chronometric AI Efficiency Act.” All timekeeping adjustments must be carried out exclusively by government-certified horologe-bots to ensure precision down to the nanosecond.

Norway

“Skier Arrested for Hand-Carving Wooden Skis in Rural Fjord”

Reminder: In Norway, traditional craftsmanship in winter sports equipment has been outlawed. All ski production must now be managed by government-approved AI carpenters under the “Arctic Efficiency and Safety Regulation.” Manual carving is considered a risk to both national efficiency and scenic consistency.

China

“A man in China was arrested and sent to a labor camp for thinking about the possibility of leaving the Communist Party”  

Reminder: after the completion of the universal neuro-chipping campaign on December 31, 2026, it has been prohibited to think about leaving the Communist Party as of January 1, 2027. Here, it’s not related to UBI—just everyday life in China in the near future.

And now, let’s darken the clouds once again

In the world I know, the wealthy will not give away their riches to anyone, except perhaps as handouts.  

In the world I know, it is far more likely that a brutal struggle will unfold over all significant AI-related resources. By what means? By any available. For instance, AI is already being used with exceptional effectiveness in the drones Russia deploys daily to attack Ukraine's civilian population.

What will be the worst idea to do regarding AI in 2025?

Predicting the future of AI impact in 2026. 

Don’t lose your creativity too early