Given my outlook for 2025 and what’s coming in high tech, it seems appropriate to kick off the year with the topic of agentic artificial intelligence (AI).
I’ve been writing about this quite a lot – both in the context of AI as well as the cross-section between AI and blockchain technology.
Agentic AI, as the name implies, is when an AI has agency. It has the ability and empowerment to take action and complete assigned tasks.
It might feel funny to think about a bunch of software code having agency in a similar way that we do in our own lives. But that’s precisely what’s so disruptive and important about this technology.
As I wrote in my 2025 outlook issue, 2023 was the year of the first wave of large language models (LLM), which were manifested in the form of a chatbot like ChatGPT. Think of the text-to-text LLMs – text in, text out.
In 2024, the industry leaned into generative AI technology, with a focus on developing multimodal models – LLMs capable of ingesting not just text but also images, video, data, audio, software code, and even real-time video feeds from the real world. We first started covering this shift in December 2023 in Outer Limits – A New Entrant Just Leapt Ahead of OpenAIs ChatGPT.
Multimodal models dramatically improved the utility of the LLMs, making them useful not just for generating answers to inquiries… but for being able to ingest a wide range of content, analyze that information, and produce useful outputs.
Material advancements were made in software code analysis and production. Legal applications of AI like electronic discovery, drafting of legal agreements, and contract analysis and synthesis found quick adoption.
The insurance industry has put the technology to use widely when it comes to underwriting, which has been a major productivity boost. And the impact of the latest forms of generative AI on the life sciences industry has been nothing short of profound.
But for any of us who don’t work in an industry or for a company that has been proactive in putting this incredible technology to work, it may not feel like “it” is happening yet.
And if you haven’t had the pleasure of having a Tesla drive you around hands-free for a hundred stress-free miles, you might think that self-driving cars are a long way away.
That line of thinking couldn’t be further from reality.
Handing Over Agency
Consider this: Google’s Gemini large language model is already a profound change in how the company searches.
If you’ve seen something like this on a recent Google search, you’re using generative AI:
Source: Google
Google has already rolled out its generative AI search in more than 100 countries, which now reach more than 1 billion people every month.
That’s right, more than 1 billion people are already using Google’s generative AI. Please don’t let anyone tell you that AI hasn’t experienced mass adoption yet.
We’d normally think of December and the holidays as a slow time in business, as everyone winds down the year, slows down on business, and spends more time with family and friends. That’s usually the case…
But last month was different.
Google DeepMind announced that it released Gemini 2.0, which is its first agentic AI model.
It builds on the multimodal capabilities of Gemini 1.5 (which we first covered the month of its launch, right here in Outer Limits – Google’s Dystopia in February 2024)…
This latest release of Gemini adds agentic technology, enabling Gemini to reason better, think about the number of steps it needs to take to accomplish a task and act on our behalf.
Here’s another way to think about it. I first made this distinction in July in The Bleeding Edge – The Agentic AI Undercurrent.
First-wave LLM technology utilized a zero-shot response approach. When we use something like ChatGPT, we give it a prompt, and then it returns a complete response. The response is based on the information from our prompt, along with its pre-trained limited knowledge, and returned in a matter of seconds.
By way of example, here’s a zero-shot or non-agentic approach – for a large language model like ChatGPT – to writing a paper…
We – the user – prompt the AI, which draws on its knowledge base from its training to produce an output.
If the output isn’t what we are looking for, we simply edit and revise our prompt, and we try again.
To state the obvious, it’s us, the user, that is doing the iteration.
The agentic workflow is entirely different. When we’re working with an AI agent (or agentic AI), it’s the AI that is doing the iteration. It has full agency to do all the work to produce what it believes is the optimal result.
And rather than a zero-shot response produced in seconds, it may take minutes or even hours to complete a given task.
I used this example in the past, but if an agentic AI was tasked with writing a paper based on a prompt, it would follow a series of steps, like the ones I have outlined below:
- Understand the user prompt/task
- Determine if any specific research is needed to complete the task
- Write a draft of the paper
- Review the draft as a critic
- Determine what needs to be improved in the paper
- Perform additional research if necessary
- Write a new draft of the paper
- Review paper again
- Write final paper
Again, this is a very different process than that of a ChatGPT experience. It should feel like the same approach we’d take if we were doing the work ourselves. It should feel that way, as an agentic workflow is very similar to human problem-solving.
An AI agent is different from an LLM though. It is given the agency to use an LLM for some of its workflow where necessary.
A simple way to think about it is that the AI agent is in charge, it’s the boss.
And it uses other forms of AI or software tools to complete the task that it was assigned.
In October 2024, we noted that agentic AIs were ready for computer use, in The Bleeding Edge – AI is Ready for Computer Use…
But that’s really just the start. Agentic AIs will be empowered with a speech generator, the ability to make phone calls, and software that can be directed to interact with websites. In October, an AI managed to create its own meme coin, command its own digital wallet, and become a millionaire in 11 days (The Bleeding Edge – The First Millionaire AI).
All of this is the reason that December wasn’t slow.
All the major players, and some of the small ones, see it all converging…
And they are racing to reach artificial general intelligence (AGI), which I’ve predicted that we’ll see by 2026.
Giving AI agency is a critical step towards achieving that goal.
The Largest AI Trend in 2025
Agentic AI will be the largest trend in AI this year.
Most either don’t understand or have discounted the importance of this AI development.
Agentic AI changes everything.
The ability of an AI to reason, solve problems that require multiple steps, and interact with the real world, is where the productivity improvements will be felt on a global scale.
And what does that mean? To be able to interact with the real world…
Think about an AI navigating and interacting with websites on our behalf. An easy example would be any kind of e-commerce application like booking and paying for flights, hotels, and rental cars. Or perhaps choosing recipes for the week and ordering all the food needed – “Scheduled for home delivery!”
Or how about an agentic AI manifested in a robotic form, capable of interacting with physical objects? That might look like a robotic dog tasked with securing and monitoring a perimeter or scoping out a dangerous location.
It will come in the form of a robotic arm in industrial and logistics settings. And it will also manifest itself in a general-purpose format like a humanoid robot – greeting the grocery delivery cart (also autonomous) and beginning basic meal prep. “Dinner served at 6 p.m.!”
Google’s Gemini 2.0 isn’t widely available yet. Access is quite limited, but there’s a ton of chatter and excitement about it.
Gemini 2.0 will be incorporated widely into search in the weeks ahead, so we’ll soon start to see the difference.
What Google has shared publicly though already shows some impressive improvements over the past two Gemini models.
But regardless of how impressive those improvements are, it’s important to emphasize that what we should be monitoring closely is the pace of technological improvement.
It’s not so important that the tech isn’t perfect yet.
It’s how fast it’s getting better.
Google Pushes the Benchmarks
Source: Google DeepMind
Gemini 2.0 is showing some very impressive jumps in software coding.
Software developers have been one of the first major sectors to widely adopt generative AI, with about 92% of all software developers already using generative AI tools. Again, don’t let anyone tell you that AI isn’t being widely used yet.
The latest version is also showing a big jump in math performance. Math has been one of the harder tasks to solve because it requires reasoning, and reasoning is critical for agentic AI to be useful, as well as for making the jump to AGI.
But what’s missing from the chart is actually one of the most important metrics: The WebVoyager benchmark, which measures how successfully an AI agent completes tasks in the real world on the internet. Gemini 2.0 achieved a score of 83.5%, which is best in class.
Again, the productivity implications of this are profound.
And this is leading to something that we’ll see later this year: a universal AI assistant.
Google calls this initiative Project Astra. Project Astra has already enabled the latest Gemini model to have 10 minutes of memory of its chats with humans, and it even can remember past conversations. Just imagine how useful that could be.
This is the future. Anyone with a computer or a smartphone will have their own personalized AI agents at their beck and call, saving us hours of work a day. And these services won’t be expensive either. It’s software, so it scales quickly and cheaply.
Pricing will be set based on inference costs plus margin (computational costs), or in some cases, it will be free (when the model is driven by data collection and advertising revenues, think Google’s suite of services).
And not surprisingly, the Project Astra team is already testing out this agentic AI technology on smart glasses, something I’ve long predicted.
The ultimate user interface is hands-free, our voices, and – what is more convenient than a screen we peck away at with our opposable thumbs – than one right in front of our eyes (lenses)?
It’s happening now. This year.
The future is agentic, and it’s personal.
We’re going to all feel like kings and queens with ridiculously useful assistants at the tip of our fingers, painlessly sorting out all our mundane tasks that have historically sucked up far too much of our time and happiness.
We’re going to welcome our new agentic personalized AIs into our lives without hesitation.
“They” will understand us and our needs. And they will remove friction in our lives, freeing up time for us to do the things we’d rather be doing.
Get ready for an incredible year ahead.
Dividend Stocks for a Richer Retirement [sponsor]
These may be the most exciting dividend stocks on the market right now.
They pay big dividend yields... have strong fundamentals that can add stability to your portfolio... and they've tripled, quadrupled, even sextupled the market over the past 2 years.
And each still has plenty of room to climb.
Whether you're already enjoying your retirement or you still have years to go, these are stocks worth checking out right away.
Download 5 Dividend Stocks to Include in Your Retirement Strategy report. Today, it's absolutely free.
