Ubiquitous, more intelligent AI agents in 2025

Industry observers are predicting shifts in the way AI is deployed in 2025. 

In a list of 2025 predictions, Pure Storage talked about AI becoming more ubiquitous, naming the phenomenon Industrial AI. "Market observers estimate that most GPUs deployed are currently severely underutilised. Additionally, the majority of GPUs are deployed in a handful of companies, including the hyperscalers, with very few in private enterprises," the company stated. 

"This will shift in 2025, as enterprises bring much of the AI capability in-house to extract even more value out of their data and 'industrialise' AI. We call this Industrial AI, which brings its own set of challenges including governance – specifically around how to train the models with proprietary data that needs to be kept confidential even between departments. Agentic AI and large quantitative models (LQMs) will play a key role in this wave."

Whereas a large language model (LLM) stores a vast amount of language-related data, LQMs store large volumes of numbers as data sets.

What will be even more significant is agentics. Rahul Yadav, Chief Technical Officer, Milestone Systems said: "Unlike traditional AI systems that follow prescribed steps, AI agents are autonomous systems capable of understanding contexts, making decisions, and taking actions independently. 

"These agents — similar to but far more sophisticated than today’s chatbots — use generative, training-based approaches rather than deterministic programming. By 2025, we’ll see these agents emerging across different products and services, from video analytics to automated security responses.

"The real power of these agents lies in their ability to reason and adapt. Unlike traditional software that needs explicit programming for every scenario, these systems can understand context and make nuanced decisions. This capability will transform everything from access control to emergency response, creating more intelligent and responsive security environments," Yadav said.

Source: Salesforce. Barfield.

Salesforce has a similar definition for agents. "Unlike chatbots and copilots, AI agents can autonomously navigate tasks and make real-time decisions directly in the flow of work — moving from mere assistance to taking action based on live data and context, marking a major step forward in enterprise AI," said Gavin Barfield, VP & CTO, Solutions, Salesforce ASEAN.

"In 2025, purpose-driven AI agents designed to address specific workflow needs and provide measurable benefits will help organisations move beyond experimentation to achieve tangible outcomes. For this to happen, generative AI needs to be grounded in the right data and delivered in the flow of work to offer meaningful impact."

"While we expect agentic AI to only become mainstream from 2026 onwards, agentic systems will change the way AI is used for decision making in enterprises next year," Pure Storage said. 

CyberArk agreed, saying that AI agents, intelligent, purpose-built tools that can make decisions on behalf of humans for specific tasks, will proliferate and mature in 2025. "We expect to see more AI agents perform specific tasks with high proficiency, enabling more tailored and robust AI applications," said Jeffrey Kok, VP of Solution Engineers for Asia Pacific and Japan at CyberArk.

"As AI systems mature, we will see an increase in AI brokers: intermediaries that combine various AI agents to deliver more comprehensive, versatile solutions. For example, Apple Intelligence could support AI agents from other platforms such as Google, Meta and others."

Qlik has also commented on the agentic AI trend. In a discussion on AI trends in 2025 the company said: "Now that AI has started to surpass human performance across benchmarks including image classification, visual reasoning, and English understanding, deploying agent-based systems that can independently execute tasks and adapt to feedback is both feasible and critical in unlocking economic value."

"Fuelled by advanced AI and machine learning (ML) models, AI agents have the capacity to go far beyond generative AI’s ability to generate content and answer questions. These goal- and action- oriented agents can develop and execute an action plan to achieve the business goals set for them. Such capability allows organisations to augment the human workforce with virtual workers possessing the 'smarts' to respond to plain language prompts and event triggers, reason through complex problems and processes, retain and reflect on outcomes to adjust and improve, and recommend and take actions," said Jess O’Reilly, Area VP, Asia, at UiPath. 

Source: UiPath. O'Reilly.

"Equipped with real-time data, context, and the right demand and behavioural predictive models, agents can either support human representatives in providing one-to-one service—or use their own conversational abilities to provide it themselves." 

O'Reilly said 2025 will set the foundation for growth in agentic AI, with early adoption of agentic use cases and investment to build out an orchestrated agentic ecosystem across the enterprise. 

"Enterprises will look at prioritising orchestration capabilities for coordinating tasks, managing workflows, and optimising operations across diverse enterprise technologies and systems. For instance, UiPath is focused on empowering agents to orchestrate processes across the entire enterprise application ecosystem, instead of targeting singular platforms, to automate at scale.

"There will also be a growing expectation for orchestration to support multiple agents, whether they are working independently or collaboratively, and to integrate their decisions and actions into coherent, well-orchestrated sequences." 

"By 2025, AI adoption will continue to be fuelled by agentic AI developments. For example, UiPath’s Autopilot for developers has reduced automation time by up to 75%, while Autopilot for tester has eliminated 50% of manual testing. And that’s just copilots—providers are increasingly leveraging AI throughout their products to expand capabilities, boost performance, and lower usage barriers," O'Reilly shared. 

Source: NVIDIA. Briski.

Orchestration was also the emphasis for successful agentic AI in Kari Briski's predictions. "Enterprises are set to have a slew of AI agents, which are semiautonomous, trained models that work across internal networks to help with customer service, human resources, data security and more. To maximise these efficiencies, expect to see a rise in AI orchestrators that work across numerous agents to seamlessly route human enquiries and interpret collective results to recommend and take actions for users," said Briski, who is VP of Generative AI Software for NVIDIA.

"These orchestrators will have access to deeper content understanding, multilingual capabilities and fluency with multiple data types, ranging from PDFs to video streams. Powered by self-learning data flywheels, AI orchestrators will continuously refine business-specific insights. For instance, in manufacturing, an AI orchestrator could optimise supply chains by analysing real-time data and making recommendations on production schedules and supplier negotiations."

"Looking ahead to the next decade, I anticipate a future where AI agents will play an active role and assist in tasks like scheduling appointments, making purchases, and paying bills, allowing us to step away from our screens," said Jay Jenkins, CTO, Akamai Technologies APJ.

"By 2025, we will start to witness the initial phases of this transformation. The chatbots we’ve become familiar with will develop into basic AI agents capable of performing simple tasks instead of merely guiding users through menus. For example, rather than just helping you navigate the process of booking an appointment with your healthcare provider, these agents could potentially handle it directly, offering you available time slots without any extra effort on your part. 

"This shift will not only redefine convenience but also free us to focus on what truly matters, heralding a new era of effortless living powered by AI."

Chris Connolly, Solutions Engineering Lead for Communications, APJ at Twilio, predicted a similar evolution path for chatbots. "Most chatbots today serve as mere stopgaps, deflecting issues until a human agent steps in to provide real solutions. As they operate based on straightforward rules and predefined content, chatbots struggle to generate responses that truly simulate human engagement. 

"In addition, chatbots trained on static resources like FAQs or manuals offer generic answers but often miss the opportunity to deliver truly individualised responses tailored to each customer," he pointed out.

"2025 will mark a turning point in front-line communications as the focus shifts towards eliminating friction in such chatbot engagements. Brands are already adopting conversational AI systems capable of better identifying user intent and generating sentences that mimic the nuances of human conversations. Conversational AI can engage with customers, reference previous interactions, and respond in a way that feels more dynamic and natural.

"Beyond conversational AI, brands are also realising the potential of intelligent AI agents that can offer/upsell products and services, take action on customer issues, make decisions within constraints, and operate across communication channels. These agents—which are not confined to chat windows—have an in-depth understanding of consumer preferences collected over time, past purchases, and previous customer interactions such as login issues or unresolved concerns. They can tailor their communication style to a specific context across channels. AI agents can even be triggered based on an event in the customer journey," Connolly said.

"The holy grail is to make these AI agents invisible yet highly effective, creating a customer service experience that feels natural, effortless and trusted. Contextual data can help turn this into reality, empowering AI agents to better anticipate consumer needs and resolve issues quickly and eliminate the hassle of requiring customers to repeat themselves."

According to Qlik, some trends influencing the successful rollout of agent systems include:

Multi-agent architectures

"Just as there are competing cloud environments and AI foundation models, expect to see multiple agentic architectures co-existing. Interoperability and avoiding vendor lock-in will be critical to realising the full potential of agentic reach and value. Some agents will be good at data integration, others at schema cleaning, text-to-SQL generation, automation, or building dashboards. Over time, these agents will learn to interact with one another," Qlik said.

"But humans must stay in the loop, or at least 'over the loop', for surveillance and governance."

Process intelligence and automation

Garbage in, garbage out. Qlik said: "Bad processes that have been automated are still bad processes. In a world of autonomous agents, it’s vital that process flows are understandable, and don’t look like spaghetti. Use process mining and analytics to optimise what workflows should look like. This will act as a highway for agents. Automations are then the transportation vehicle of the agents on that highway, safely connecting applications, facilitating them talking to one another, driving agent-to-agent interactions and actions."

Real-time data

"Up-to-date data is critical to trust agents. A customer service agent can’t make a decision or give advice based on stale inventory data. Real-time data not only gives an immediacy advantage but reduces the chance of it being stale, irrelevant or incorrect. The need for real-time is leading to a substantial evolution in architectures," Qlik noted.

"Technically, we’re reaching an inflection point where ingestion and transformation of data can be done in real-time, and with hybrid transactional and analytical data can be stored and processed in the same place."

Rethinking the role of applications

"A world with agents forces us to rethink applications. Sometimes, we don’t need applications, as agents can fetch the answers we are looking for. Other times we will want to buy prepackaged applications for expediency and domain-specific logic," Qlik said.

"The combination of text-to-action, large context windows, and agents will also enable us to build more apps in-house. As applications become more dynamic and intelligent, they’ll morph into alignment with our changing needs and learn from new data to deliver more personalised, predictive, and context-aware experiences."

Governance

The rise of agentic AI also raises ethical and regulatory challenges, cautioned Ying Shaowei, Chief Scientist, NCS. "The capacity for autonomous decision-making introduces potential risks, from unintended actions to biases in the agent’s learning process. As organisations consider agentic AI, robust governance frameworks will be essential to prevent undesirable outcomes," he said.


Explore

There are other comments about AI in general as well as about cybersecurity and AI in other posts.

Hashtag: #2025Predictions