Good morning. Over the past two months or so, I’ve been collecting predictions from people across the AI industry about what 2025 might look like. 

So, over the next week, we’ll be taking a bit of a departure from the AI news of the day to instead take a deep look, both into 2025 (and back at 2024) in a progress and predictions series that aims to examine how far AI has come, and where it might be headed. Had a lot of fun putting this together.

We’re kicking things off today with a look into one aspect of technical progress, crossed with the question of enterprise adoption. The consensus here is that 2025 will become known as the ‘year of the AI agent.’ 

— Ian Krietzberg, Editor-in-Chief, The Deep View

Make 2025 your Best Year With This $399 FREE AI Masterclass

If you don’t become an AI-enabled professional in 2025, you will:

This 3-hour power-packed workshop will teach you 25+ AI Tools, make you a master of prompting & talk about hacks, strategies & secrets that only the top 1% know of.

Save your seat now (Offer valid for 24 hours only ⏰)

Here’s why you can’t miss this:

✅ Learn 30+ cutting-edge AI tools to completely transform the way you work, market, and lead.

✅ Automate your workflows and reclaim 20+ hours a week (imagine no more burnout).

✅ Create an AI-powered version of YOU—a digital clone that handles the tedious work while you focus on what truly matters.

So don’t think twice, you have NOTHING to lose. Consider it a $0 investment that will yield value worth millions.

⏲️ Time: 10 AM EST (Tomorrow)

Register now or miss your golden chance

But first … A look back at 2024 and the rise of the AI chatbot

Before we get to the agents and what might lie ahead, we have to start with chatbots, and the language models they’re built on. 

The rise of generative AI: In the beginning, there was nothing. Then, there was ChatGPT. Ok, well, that’s not strictly true, but the launch of ChatGPT — a generative interface built on a Large Language Model (LLM) — is regarded as the moment that sparked the current AI boom. 

Importantly, it made AI technology — which, beforehand, was an under-the-hood kind of technology — visible. People could interact with it. And that’s what changed everything. That moment came at the tail end of 2022, making 2023 the year of the chatbot race and 2024 the year of the chatbot (OpenAI chief Sam Altman said in December that ChatGPT had reached 300 million weekly active users). 

Corporate spending on generative AI, according to a report last year, spiked from $2.3 billion in 2023 to $13.8 billion in 2024, a 500% increase. The top five use cases here involved code generation, support chatbots, enterprise search and retrieval, data extraction and summarization and meeting summarization. 

Even as the bulk of AI projects in the enterprise remained unable to get off the ground, last year was a year of serious contemplation for corporations about the opportunities afforded them by generative AI. As Microsoft told me, the experimentation phase has largely concluded; now, people are working on the application phase. 

And that leads us to agents. 

Never Prep for a Sales Meeting Again

Bounti’s AI will save you 10+ hours/week of painful account research and meeting prep, empowering you and your team to sell more effectively through every channel and stage of the funnel.

Deliver the right message to your ideal buyer—driving real engagement at a fraction of the cost of other solutions.

Get started in minutes—not weeks—for nothing

The agents are coming

As the tech becomes more integrated and normalized, expectations around application have shifted; chatbots are cool, but developers — particularly in enterprise offerings — have become increasingly interested in the evolution of the chatbot, which is something the industry has dubbed the “AI Agent.” 

Microsoft, Google and Nvidia have all begun positioning and pushing so-called agentic offerings. 

The idea here is simple in concept; where a chatbot produces output in the form of imagery, textual or video content, an AI agent — built, of course, on the same LLM-based architecture — would complete actions. Though each company has a different definition of an agent, according to IBM, the term “refers to a system or program that is capable of autonomously performing tasks on behalf of a user or another system by designing its workflow and utilizing available tools.”

Deloitte expects that 25% of the enterprises currently using generative AI will deploy AI agents next year, a number that will rise to 50% by 2027. Deloitte expects the capabilities of these systems to vastly improve throughout next year, a boon to their adoption and deployment. 

It is worth noting that issues of bias and hallucination in LLMs — in addition to their massive energy footprint — have not been solved; action-taking generative AI therefore poses risks of, not mistaken content, but mistaken actions, which would presumably have much more significant consequences. 

Still, industry experts that I spoke with think that 2025 will see the rise of so-called agentic software. 

Dr. John Licato, an associate professor of computer science at the University of South Florida and the founder/director of its Advancing Machine and Human Reasoning lab, told me that the biggest development in LLMs thus far involves the growth of agents. 

Dr. John Bates, a former professor who is now the CEO of SER, told me that, with the “information tsunami” expected to get worse and worse, agentic AI might offer a solution. 

Uzi Dvir, CIO at WalkMe, said that, as we get into 2025, the “agent wars are heating up.” 

A look back at related stories we’ve done in the past year: 

It’s an impression shared by Hyperscience CEO Andrew Joiner, who said that the idea of human-in-the-loop technology will begin changing next year. 

This push toward a more autonomous environment is one that many expect will translate to enterprise adoption, under the expectation that it will enable clearer and more actionable return on the massive investment of AI. 

Hyperscience’s CTO Brian Weiss expects that, next year, generative AI will no longer be a solution in search of a problem, something we’ve talked about several times in the past year — researcher and software engineer Molly White noted last year that Large Language Models (LLMs) “do a poor job of much of what people try to do with them, they can't do the things their creators claim they one day might and many of the things they are well suited to do may not be altogether that beneficial.”

Weiss said that the adoption approach toward generative AI has thus far gone backwards, with industries exploring with the tech by playing with its capabilities first, rather than identifying solutions to specific problems. This, he said, will change in 2025. 

But Galileo co-founder Yash Sheth told me that, while integration will continue ramping up in the enterprise, the process won’t be a smooth one, and it won’t be one-size-fits-all. 

“The industry has been led to believe that implementing AI into the enterprise is extremely easy,” he said. “In 2025, we’ll see the realization that there isn’t an ‘easy button’ and an evolution beyond simple interfaces (ChatGPT chatbot → RAG + Fine-Tuning) and into sophisticated implementations.” 

Still, Sheth said that the “world is moving towards just extracting more ROI from generative AI.”

The “whole trend towards agents is nothing but to get more ROI from generative AI systems. Automation is the end goal. You need to automate a lot of things using AI, where humans are involved and without true actions being taken by AI, it's not going to be full automation. Agentic systems are going to actually lead us to that automation and that ROI that we expect from generative AI.” 

As a result of this ever-increasing automation, computer scientist and AI expert Dr. Srinivas Mukkamala expects that it will start to become “radically difficult” for high-paying software developers to find a job in 2025 and beyond.

A look back at related stories we’ve done in the past year: 

Get ready for my own predictions …

In this conversation about the move to agentic AI and a more widespread integration of the tech by enterprises, I expect the road to perhaps be a little less rosy than it’s been thus far. Enterprises have had two years now to mess around with generative AI. And they’ve seen that hallucinations and algorithmic bias are a real problem, neither of which are going to go away. 

They’ve also been faced directly with the massive price tags associated with this tech, price tags that would be justifiable if things, well, worked — as we wrote about last year, some customers have grown so frustrated with the usability and cost of Microsoft’s flagship Copilot product that they have paused their subscriptions to it. 

So I would imagine that, as opposed to 2023, when every executive on the planet was anxious to say the word “AI” to appear with the times, we’ll see more caution in 2025. I think businesses are aware that there are gains to be made with AI, but maybe in less obvious ways — perhaps an email assistant isn’t worth the hassle and the cost, but maybe agentic AI solutions to track and cut down on energy usage could result in cost-savings that make an AI investment more than worthwhile. 

As part of that, I do think we will start to see a push away from single models and chatbots; chatbots aren’t enormous time-savers for most people. This is what’s behind the push for agents. And so, because of this, I do think we’ll see attempts at agents. I think this push is premature, however, and will result in a few enormously costly hallucinations that will make businesses think twice about deploying agentic solutions. 

On the model front, there’s been a gradual understanding that models are fine, but systems are far more useful. In the enterprise especially, looking at generative AI as a more holistic, more traditional digital system enables better cybersecurity protections as well as better results (think big models and small models integrated together in a stack). I expect we’ll start to see enterprises pushing in this direction (whether you want to call that an ‘agent’ depends on which company’s definition you choose to stand by).

And, on the agentic front, again, I think the technology is premature for generative AI to jump into decision-making and action-taking, and I think it’ll result in costly errors that will spook customers (just think of all the ways a hallucination could screw up buying airplane tickets without careful oversight). I do, however, expect that all the major developers will launch clear iterations of their own versions of agentic AI — which will remain bounded and limited by the architecture; likely, these products will be more costly than current solutions, and we’ll just have to see if businesses will be interested in paying for that.

And that brings me to my final related point here — so far, the developers, through a combination of venture capital dollars and Big Tech cash infusions (in the billions) have been able to subsidize the cost of using generative AI as part of their desperate push to win users over to the tech. This will not continue in perpetuity. 

I fully expect that, in 2025, we will see major developers move from talking about increasing subscription prices, to actually increasing baseline subscription prices, something that has already started to happen. OpenAI in December unveiled a $200/month Pro tier (though it did not axe its $20/month Plus tier). I would expect these significant price increases to become a norm across the industry; these businesses are burning cash at a rate that is difficult to comprehend, and with adoption set to hit higher levels in 2025, it seems like a good time to test the willingness of the market to cough up. (OpenAI, according to Sam Altman, is losing money even on the $200/month tier; maybe things will go higher still).

I think this will push a large swath of people and smaller businesses away from the product. 

The last thing I’ll say here is that, enterprises aside, the push into agentic AI will usher forth a moment of reckoning for modern society; will ordinary people accept and pay for the integration of systems that (unreliably, and at great cost) just do things for them? Or will they deem it ill-fitted to daily life? In this way, 2025 will be a telling year for the future, not just of AI, but of civilization. Will things be rejected? Will they be accepted with clear guardrails and limitations? Or will they be uncritically adopted? I can’t predict how this will play out, but it will be interesting to watch, with massive implications across a wide spectrum of ethical, economic and environmental issues.

🤔 Your thought process:

Selected Image 1 (Left):

Selected Image 2 (Right):

If you want to get in front of an audience of 200,000+ developers, business leaders and tech enthusiasts, get in touch with us here.

💭 A poll before you go

Thanks for reading today’s edition of The Deep View!

We’ll see you in the next one.