What Business Leaders Should Know about AI Trends in 2025
As we usher in 2025, artificial intelligence (AI) continues to dominate many business and technological developments. Although generative AI has led the charge in this global race toward AI modernization, the latest trends in AI research, technology, and policies encompass much more than that.
Recap: AI-enabled automation
In May last year, we released a trend insight article about AI, API, and automation.
The appetite for AI-enabled automation has continued into 2025, with agentic AI identified by Gartner as the top strategic trend to look out for. There is reason to hop onto the hype: market trends point towards improved AI capabilities, especially for multimodality, reasoning, and decision-making. There is also reason to tread carefully, as researchers and policymakers work to understand the behavior and effects of AI.
Following the push for more automation, here are Orkes’ top four observations and predictions about AI for 2025 and beyond.
1. AI will continue to redefine human-computer interactions
AI and generative AI will continue to be the top drivers in redefining human-computer interactions. In the previous decade, voice assistants like Siri and Alexa promised a new way to interact with technology via voice. Likewise, immersive technology like AR and VR had potential for melding physical and digital spaces in new ways (recall: Meta’s foray into the Metaverse). While such technologies have always been augmented by AI and machine learning, the rise of LLMs will unlock even more ways to interact with the digital world.
With LLM-powered chatbots like ChatGPT and Claude, the possibility of interacting with computer systems via natural language (spoken or written), rather than pre-fixed buttons and commands, is fast becoming day-to-day reality.
Today, many digital customer experiences offer AI-powered chatbots that can intelligently field queries and resolve customer issues before escalating them to a human agent. Meanwhile, tools like Perplexity AI is transforming how we search for information online.
The opportunities for redefining human-computer interaction don ’t just stop there. AI is also lowering the technical barrier to providing multimodal experiences with native text-to-speech, speech-to-text, computer vision, or image/video recognition and generation capabilities.
Project Mariner promises “a new way to use your browser” with multimodal interactivity, reasoning capabilities, and task automation features; and robots with embodied multimodal capabilities (like PaLM-E) can dynamically interact with the physical environment and humans based on natural language instructions.
In 2025, we expect to see more projects or product offerings that will use AI to enhance digital and physical experiences in novel ways.
What does this mean for businesses?
Growth & business value
Like the first digital boom in the 1980s, AI today presents an ever-growing opportunity for companies to redefine how they can provide goods, services, and experiences. The use cases for reshaping the digital landscape range across generative AI virtual assistants, multimodal UI, agentic AI, AI avatars, and embedded or edge AI. An analysis by PwC estimates that AI stands to contribute $15.7 trillion to the global economy by 2030.
However, instead of rushing to tack on AI to any product, businesses should carefully weigh how such AI technologies can truly elevate their core business value before implementing them.
2. AI capabilities will keep growing, even if at a slower pace
AI performance may have slowed in recent months, but it has not hit saturation point yet. In general, AI performance is affected by three key factors: compute, data, and algorithms/neural network architecture. As recent years have demonstrated, AI capabilities have exploded exponentially, outperforming or nearing human benchmark performance in various skills.
Yet within this short timeframe, AI firms and researchers have also run into a number of roadblocks. First: model size. While 2023 has been the year of large language models with billions of parameters, the leaps in such large model performances have seemingly slowed down towards the end of 2024. Second: quality training data. Researchers have estimated that we may hit the limit of public human-generated data within the next 1-7 years, which could signal the ceiling for AI progress without curated private data or synthetic data.
The good news is model size or data is not the only way to go, when it comes to improving AI performance. Outsized leaps in the performance of smaller models, and AI techniques like transfer learning all go a long way in boosting performance.
What’s more, new techniques are plenty: Anthropic recently shared how using Contextual Retrieval in an RAG system enhances retrieval performance. Meanwhile, novel architecture like state space models (SSMs) and Fourier analysis networks (FANs) are poised to propel improved reasoning capabilities and other deep learning applications.
In 2025, we expect to see (1) a greater focus on optimizing domain-specialized or smaller models, (2) improvements in reasoning capabilities, multimodality performance, and context windows/sizes, and (3) more AI techniques becoming feasible in real-world use.
What does this mean for businesses?
Viable opportunities & use cases
We may soon be breaking new ground, where AI can complete more advanced tasks that involve complex reasoning and analytical thinking. Furthermore, with smaller or domain-specialized models demonstrating performance comparable to large models, it will become much more affordable to leverage the right AI tool for any enterprise case.
Businesses should keep an eye out for the latest developments that will lower costs and unlock improved AI capabilities. The time to use AI for decision intelligence, big data analysis, and AI agents is soon within reach.
3. Autonomous AI is on the horizon
As one of the most promising AI developments, agentic AI will be the talk of 2025. The concept of AI agents isn’t new. Early agents were rule-based systems that fell short of handling complex or dynamic situations. With the rise of LLMs enabling context-aware interactions, AI agents are once again back on the discussion table.
Today, the state of the market is still nascent—many agents that are currently available are large language models (LLM) chatbots with additional abilities, like a copilot that generates functional frontend code from a website idea or Google’s agentic Deep Research feature on Gemini 2.0. As memory and state management for agents mature, AI agents can soon persist in complex, long-running tasks, like handling evolving project requirements or remembering repeat clients.
2025 will be the year of agentic product releases, as both tech giants and startups contend for market share. Likewise, we expect to see more agent-building frameworks and platforms emerging to support this burgeoning ecosystem. While the road to fully autonomous AI (both digital and physical) remains a long way to go, the growing capabilities of AI mean that decision intelligence systems and tool-use AI agents could become viable in the next few years.
The benefits and use cases of AI agents, with their autonomous and emergent behavior, are quite clear. But the risks are less so. LLMs are prone to hallucination, and ongoing research indicates that AI models, especially LLMs, can do or say unexpected things without being instructed—especially unethical actions like deception, manipulation, or alignment faking. This takes on a more severe tone when autonomy comes into play.
What does this mean for businesses?
Time to invest and take up new ventures
With all the excitement around agentic AI, it may be difficult to discern its true abilities from the hype. Nevertheless, early adopters stand to gain a competitive edge by investing in AI agents—lowered costs, improved productivity, or greater innovation. As with any new venture, companies can take steps to prepare by partnering with AI experts and starting small with a proof of concept or minimum viable product.
4. AI governance and data privacy will play catch-up
All these leaps in AI translate to a game of catch-up for governance, compliance, and data privacy. A 2024 survey revealed that less than half of enterprises are compliant with existing AI regulations or actively working towards compliance. Furthermore, existing regulations for AI worldwide are uneven or spotty, with the EU AI Act being the first notable law taking effect. We expect to see more countries across the globe follow suit in the coming year.
On the private market side, we expect to see more AI governance startups and features that help monitor and audit AI risks. Likewise, LLM providers like OpenAI and Anthropic are expected to strengthen their enterprise security and privacy measures.
Ongoing conversations highlight data leaks as a key concern—if an AI model powering a platform has access to all data, how will it know to constrain a user’s access to their own data and not others? This issue remains to be addressed if generative AI or agentic AI is to be adopted at the enterprise level. In past years, such conversations have taken the sidelines in favor of all the hype. However, 2025 looks to be the year when more frameworks, technologies, and regulations will come into place for safer, more responsible AI.
What does this mean for businesses?
AI literacy & due diligence
With a rapidly evolving landscape of regulations to navigate, it becomes increasingly difficult to stay abreast of these changes. Nevertheless, it is important for companies to implement compliance measures and practices for mitigating AI-related data risks. When adopting AI for internal business processes, the most prudent approach would be a gradual rollout with due diligence.
Wrap up: AI in 2025
2025 is set to be another mercurial and fast-paced year for AI developments. Besides the growing buzz about AI agents, here is the summary of the shifting winds we expect ahead:
- Productization:
More products. More startups. More giant-backed ventures. More AI projects evolving from incubating in research to market-ready products. - Scale:
Choose from a wider range of AI models as we see consolidation at the high end and democratization at the low end. - Multimodality:
The seamless transitions between text, speech, image, and video will unlock new ways to interact with AI. - Smaller models:
Small but mighty—smaller, specialized models will come to carve out their place in the AI landscape.
Orkes makes it easy to build durable and responsive enterprise applications and processes.
Orkes Conductor is an orchestration platform for automating and monitoring complexity across APIs, microservices, event-driven systems, AI/LLM, and human-involved tasks. With Conductor, you can seamlessly integrate with any LLM to bring the power of AI into your business use cases and workflows. Learn more about Conductor and how it works under the hood.
