Rose is a thought leader in Digital Journal’s Insight Forum (become a member).
If you’re even half paying attention to AI, you know 2025 is the year of the AI agent. There will be significant mainstream adoption of AI agents, similar to the “ChatGPT of 2022” event.
What are AI agents? You can think of them as digital workers. They are autonomous software programs that execute tasks towards a goal, or make decisions on your behalf.
How do I know 2025 is the year of the agent? I follow signals, and based on a number of factors, I can confidently make predictions about how big of a trend any given signal will turn into. It’s not just me, Gartner predicts that by 2028, 33% of enterprise software applications will deploy AI agents (which is up from less than 1% in 2024). Improvements in AI capabilities alone have been the biggest signal that this year will be huge for autonomous AI.
Multimodal models
Text-only models are so 2023 — now we have truly impressive multimodal models, essential to enabling agentic AI. A multimodal model is an AI system capable of processing and understanding multiple types of input data — such as text, images, audio, and video — all within a single unified architecture. Unlike earlier AI models that specialized in just one type of data (like text-only or image-only models), multimodal models can seamlessly work across different formats, understanding the relationships between them.
Think of it like a human who can simultaneously read text, look at pictures, listen to audio, and watch videos, all while comprehending how all these different types of information relate to each other.
Consider an AI agent tasked with managing a company’s social media presence. Thanks to multimodal capabilities, it can now analyze engagement metrics while simultaneously assessing image quality, evaluate video content for brand consistency, convert customer service calls into actionable insights, and even generate new visual content that matches the brand’s aesthetic.
This is just one example of how an agent can process different types of media, while understanding the relationships between them and make sophisticated decisions about how to use each format for maximum impact.
Reasoning capabilities
The quantum leap in reasoning capabilities we saw in 2024 marks a defining shift in AI development. We’ve moved past simple pattern matching and statistical prediction. This is why OpenAI’s o1 reasoning model release in December of last year was so remarkable. We’re witnessing the emergence of AI systems that can plan, execute, and course-correct without human intervention, which is a marked improvement from the rigid, rule-based automation we’re used to.
This is all possible because of advancements in models that can emulate structured human thought processes. Reasoning in an AI model refers to its ability to process information logically, make connections between concepts, and arrive at conclusions or decisions through a structured thought process.
While traditional language models focus primarily on generating or understanding text based on statistical patterns, reasoning models go a step further by using structured approaches to excel in logical problem-solving, decision-making, and understanding complex relationships between concepts.
Increased context windows
Remember when we got excited about 4K token windows in 2023? In February 2024, Gemini 1.5 gave us a 1M+ token context window. Even that feels almost antiquated now. The promise of unlimited or infinite token context windows is fundamentally transforming what AI can accomplish. What makes this particularly powerful for autonomous AI is the ability to maintain consistency across long-running tasks. Unlike earlier models that would “forget” important context after a few exchanges, modern agents can maintain coherent understanding across extended operations — making them much more reliable for autonomous decision-making and task execution.
The most significant impact is on task continuity and complexity. With larger context windows, these agents can now maintain awareness of long-term goals while executing immediate tasks, hold multiple document contents in their “memory” while working across them, track complex multi-step processes without losing sight of earlier steps or requirements, and carry forward important context from hours of previous interactions. This expanded memory capacity allows agents to operate with a level of continuity and sophistication that much more closely mirrors human cognitive abilities.
Deep learning
The evolution of deep learning architectures has been nothing short of revolutionary.
We’ve moved beyond traditional transformer models to hybrid architectures that combine the best of multiple approaches. These new systems demonstrate incredible adaptability, learning from vast datasets as well as minimal examples — something we call few-shot learning.
The models are also showing increasingly sophisticated capabilities in transfer learning, where knowledge gained in one domain can be successfully applied to another. This mimics human cognitive flexibility in ways that seemed impossible just a year ago.
Increase in investment in compute
The arms race in computational infrastructure has reached fever pitch. Major tech players and nations are pouring unprecedented resources into semiconductor fabrication and data center expansion.
We’re seeing the emergence of specialized AI chips that make previous generations look like pocket calculators. The investment goes beyond hardware to the entire ecosystem of AI compute. AI compute is the technology that enables AI systems to perform tasks like processing data, training machine learning models, and running algorithms. New cooling technologies, more efficient power management systems, and innovative data center designs are all part of this computational gold rush. The numbers are staggering: industry analysts project 2025’s AI compute investments will dwarf the combined total of the previous three years.
More signals that 2025 is going to be monumental for AI agents are significant recent product developments, and planned releases from frontier companies.
- Salesforce unveiled its suite of AI agents for enterprise, Agentforce, in September of 2024
- Models with capabilities to act like assistants such as Microsoft’s Copilot powered by Prometheus becoming more widely available to SMB enterprise January of 2024 , and Google’s NotebookLM powered by Gemini 2.0, becoming widely available to all users (18+) in June of 2024
- Google’s Project Mariner, an advanced AI agent and prototype powered by its Gemini 2.0 framework, is set to be released to the wider public in 2025
- OpenAI’s o1 reasoning model release in December 2024
- OpenAI’s rumored AI agent tool codenamed “Operator” scheduled to be released in 2025
Finally, some of the biggest names in tech, who work on this technology every day, have similarly proclaimed the rise of AI agents:
Sam Altman, CEO of OpenAI: “We believe that, in 2025, we may see the first AI agents “join the workforce” and materially change the output of companies.”
Bill Gates, co-founder of Microsoft: “Agents are not only going to change how everyone interacts with computers. They’re also going to upend the software industry, bringing about the biggest revolution in computing since we went from typing commands to tapping on icons.”
Satya Nadella, CEO of Microsoft: “SaaS applications or biz apps – the notion that business applications exist, that will probably collapse in the agent era.”
Just to be clear, the year of agents has already begun. In the enterprise space, we’re seeing AI agents that can autonomously manage entire workflow pipelines, from initiating projects to quality control and deployment. Financial institutions are already deploying autonomous agents that coordinate across trading, risk assessment, and compliance functions. In software development, we’re seeing agent swarms that can collectively architect, code, test, and deploy entire applications with minimal human oversight.
The continued breakthrough lies in agents’ ability to increasingly handle “fuzzy” objectives — turning vague instructions like “optimize our customer service process” into concrete, multi-step action plans. Particularly exciting is the emergence of agent-to-agent collaboration. Agent systems or agencies can divide complex tasks among themselves, negotiate priorities, and reconcile conflicting approaches — all while maintaining alignment with human-specified goals.
The implications for business process automation are profound. Unlike traditional automated systems that break down when encountering edge cases, these new AI agents demonstrate remarkable adaptability. They can reason through novel situations, draw on relevant past experiences and even explain their decision-making process in terms business stakeholders can understand.
This combination of autonomy and transparency is what’s driving the massive wave of AI agent adoption we’re about to witness in 2025.
How would your operations change with AI agents working around the clock?
Now is the time for business leaders to evaluate the potential for AI agents to transform their business. Start by identifying repetitive, time-consuming tasks that could benefit from automation. Explore partnerships with AI innovators and agencies that can help you implement AI solutions for maximum return on investment.
Taking small, strategic steps today can set your business on the path to success in a world increasingly being shaped by autonomous artificial intelligence.
