The 2-Z of 2025 predictions: A is for AI, part 2

AI predictions for 2025 continue below:

Agents

Source: Salesforce.
Abraham.

Autonomous agents provide opportunities for topline growth, said Sujith Abraham, Senior VP and GM, Salesforce ASEAN. "These agents will deliver on the promise of AI, succeeding where solutions such as copilots fall short," he explained.

"One limitation of earlier innovations was their siloed focus on unstructured data. For example, copilots could only act based on data within emails or presentations. This misses critical transactional context, such as customer purchase history or product details. Copilots could only see part of the customer story causing them to fall short on providing actionable insights, and businesses, in turn, missed out on opportunities to foster deeper and more compelling customer relationships that generated new revenue streams.

"Autonomous agents can significantly impact a company’s growth trajectory. Take a bank that works with thousands of business clients as an example. An initial analysis of spend may lead the bank to think that most of their customers are SMEs with small spends. But a deeper look reveals that these businesses are spreading their spend across banks. It is extremely difficult to turn the workforce around to deepen engagements with all customers. Imagine if they implement autonomous agents to maintain consistent customer engagement without constant human oversight. And agents operate 24x7 – think of how much coverage across customers is now possible. The bank ultimately increases its revenue base which might otherwise be lost to competitors."

"AI agents also allow sales teams to automatically prequalify leads before handing them to human agents," Abraham added. "This way, human agents do not waste time on unresponsive prospects, basic inquiries, or low-engagement leads, which can significantly improve productivity and bottom line."

There's business-to-business, B2B, machine-to-machine, M2M, and now agent-to-agent, A2A. "Just as organisations have employees specialised in specific functions, AI agents will soon be assigned unique roles within a network. These agents will work alongside human employees, communicate with other agents, and create new agents as business needs evolve. Each agent will have a defined function, allowing the network to handle a wide range of tasks efficiently," said Gavin Barfield, VP and CTO, Solutions, Salesforce Asean.

"In this agent network, meta-agents will be crucial, coordinating actions across other agents to keep workflows seamless. For instance, a concierge agent might interact with users, guiding them on tasks it can assist with and providing updates on task progress. An orchestration agent would assess user needs and route requests to the appropriate agent, ensuring tasks are managed effectively. This setup enables collaboration on platforms like Slack, where human employees and AI agents can interact as a unified team, enhancing responsiveness and coordination.

"This new era of agents will redefine collaboration, creating a blended environment where humans and agents work side by side to enhance productivity, improve customer experiences, and support business growth through streamlined operations." 

Automation

Mei Dent, Chief Product & Technology Officer, TeamViewer, predicted that AI will establish itself as a copilot, but said true automation remains on the horizon. "The AI conversation over the past couple of years has been plagued with the notion that AI will soon fully take over. However, 2025 will be characterised by the widespread adoption of AI as an intelligent assistant rather than a fully autonomous decision-maker," she said. 

"While AI will become deeply integrated into workflows and decision-making processes, human oversight and intervention will remain crucial, particularly in business operations. The industry will move beyond the hype of full automation to focus on practical, collaborative human-AI partnerships." 

Dent's advice to organisations is that they should focus their AI strategies on augmenting human capabilities rather than pursuing full automation. "This means designing workflows that leverage AI's strengths in processing data and generating insights while maintaining human judgment for critical decisions," she said. 

"While AI shows promise in automating workflows, both technological limitations and regulatory considerations will keep full automation at bay through 2025. Companies should invest in training programmes that help employees effectively collaborate with AI tools, focusing on areas where human expertise and AI capabilities can create synergistic outcomes."

Calibration

"Copilots can help users become more efficient and grow usage to more people who might not otherwise touch analytics. But their value is increasingly being questioned, especially as some slap on a hefty fee for this service," noted Qlik in the discussion of AI trends.

"Copilot deployments need to understand use-cases better, be more proactive to fetch anomalies, and focus on solving fewer problems with more depth and relevance."

Cost

"Cloud data efforts turned out to cost a lot more than many had expected, especially when direct-querying large amounts of data. Now every gen AI prompt costs more than a normal search query, thanks to the required investment in back-end compute and chips — a cost increasingly being passed on to users," Qlik said in a discussion of 2025 AI trends. The company also noted that new AI reasoning models would cost more.

Legacy

Source: Aicadium. Poon.

The key to success lies in blending new approaches with established AI technologies and business rules, said Phoebe Poon, VP of Product Management at Aicadium.

"Technologies like predictive analytics and computer vision have already proven their value in improving operational efficiency, reducing costs, and enabling data-driven decision-making in industries such as manufacturing, logistics, and transportation. Rather than chasing the hype, businesses can focus on integrating gen AI into these existing frameworks, accelerating innovation cycles for products that have already demonstrated measurable success," she advised.

"This strategic approach ensures that organisations not only embrace cutting-edge advancements but also deliver tangible outcomes that align with their core business objectives." 

Generative AI is abbreviated as gen AI.

Middleware

Anthony Spiteri, Regional CTO APJ, Veeam Software said demand will increase for AI middleware. "In 2025, businesses grappling with the challenge of speedier adoption of secure, responsible and efficient AI solutions will increasingly engage AI middleware companies. Middleware simplifies the adoption process by allowing different systems to communicate seamlessly, reducing the need for in-house AI expertise," he elaborated. 

"According to IDC, investments in AI and generative AI will continue to increase at a compound annual growth rate of 24% from 2023 to 2028. By leveraging third-party expertise, organisations reduce the risks associated with AI development and improve time to market."

Sustainability

"The race to AI will continue to accelerate, leading to higher power consumption levels, which will impact carbon emissions across all scopes. By 2027, AI usage alone is predicted to use as much water as all of New Zealand. In a world where net-zero goals and carbon emission reductions target loom large, companies need to find solutions that balance sustainability goals with the growth opportunities that AI brings," stated Tay Bee Kheng, President, ASEAN, Cisco.

"To achieve this balance, businesses will increasingly seek partners who can provide energy-efficient products and solutions, or help them adopt circular business models aligned with their net-zero timelines. AI itself will play a crucial role in unlocking efficiencies for businesses. 

"We see AI heralding a new era of energy networking, which combines software-defined networking capabilities, and an electric power system made up of direct current (DC) micro grids. This will enhance energy efficiency while delivering increased visibility, insights, and automation. Power over Ethernet is another innovative solution, delivering DC power to devices over copper Ethernet cabling, which eliminates the need for separate power supplies and outlets."

Tay added that advancements in materials and design processes will also contribute to balancing sustainability needs, as will takeback and reuse programmes.

Cost governance will drive sustainable practices, Qlik said. "Businesses must now factor in both the expense and what the energy requirements mean for their sustainability initiatives. Cost governance will be key to keeping track of expenses without compromising the use of models. Tactics include packaging answers to reduce query volume, separating training and inference, deploying open table formats, and using smaller distributed models for specific tasks."

Vishal Ghariwala, CTO for SUSE Asia Pacific, also noted that sustainability will influence how the AI story plays out. "AI, particularly large language models, requires substantial amounts of energy to operate. This has a significant impact on CO₂ emissions. To optimise the energy consumption of their AI workloads and reduce CO₂ emissions, companies will be adopting a range of strategies in the coming years. These include finetuning and optimising LLMs, eliminating performance bottlenecks in LLMs, selecting energy-efficient cloud providers, and scaling AI resources responsibly. Observability tools will be instrumental in enabling and guiding these efforts," he said. 

Another way to increase sustainability will be to go for a standard platform for AI operations, Ghariwala added. "There are multiple options available today to run gen AI and other types of AI workloads. Over time, we can expect most companies to create a standard operating environment for AI use cases. Such an environment will comprise a common AI platform that is highly scalable and provides common modules and services required by AI workloads such as a curated set of LLMs, data privacy and security, observability etc," he said.

"A standard operating environment also ensures consistent governance, efficient workflows, and optimised resource usage, thereby contributing to reduced CO₂ emissions."

Source: RackSpace. Banerjee.

"CIOs should consider the environmental impact of AI model training and explore data minimisation strategies to reduce the carbon footprint associated with data processing and storage," shared Hemanta Banerjee, VP of Public Cloud Data Services, Rackspace Technology.

"Implement data minimisation by focusing on high-value, relevant data for AI training and leveraging synthetic data to reduce the need for resource-intensive data processing. Explore sustainable AI development practices, such as federated learning, which can decrease resource demand by training AI models locally."

Transparency

"Recent developments, such as high-profile AI failures, increased regulatory scrutiny, and concerns over bias, copyright infringement, and environmental impact, have heightened the demand for transparency and accountability in AI deployment—like ChatGPT’s generative AI systems being scrutinised for misinformation and unintended biases," said Poon.

"Key aspects requiring transparency include the sources of training data, the decision-making processes behind AI outputs, measures to mitigate bias, privacy and security protocols, and mechanisms for error management. Moreover, responsible innovation—defined as the ethical, inclusive, and sustainable design and application of AI technologies—will be crucial. This includes reducing environmental impact, fostering fairness, ensuring that AI aligns with societal values, and promoting trust as adoption accelerates in an increasingly AI-driven world." 

"When AI models are used for decision making, the data on which it has drawn from to make those predictions need to be explained. We already see this becoming the standard in search engines that experiment with generative AI results," Poon concluded.

4th wave

Source: Salesforce. Barfield.

A new wave of AI technology will appear, Barfield said: robotics. "Beyond agents, robotics will see interactions evolve from text and voice systems to immersive experiences with physical robots and virtual avatars with lifelike, dynamic, and highly interactive engagements," he forecast.

"Picture virtual avatars powered by AI agents, with heads that move, lips that smile, and expressions that react naturally during interactions. This evolution will create more personal and engaging experiences in physical settings like a concierge in shopping malls, where customers can hold a real-time conversation with a lifelike avatar rather than typing queries into a screen.

"At the same time, physical robotics will bring AI agents into the tangible world, unlocking new opportunities in environments such as F&B. Imagine a robotic barista powered by an AI agent that can offer personalised drink recommendations based on a customer’s past orders—including details like sugar preferences—and prepare the drink instantly."

Source: NVIDIA. Fan.

Linxi (Jim) Fan, Senior Research Scientist, AI Agents, NVIDIA, also had a prediction about robots. "Robots will begin to understand arbitrary language commands. Right now, industry robots must be programmed by hand, and they don’t respond intelligently to unpredictable inputs or languages other than those programmed. Multimodal robot foundation models that incorporate vision, language and arbitrary actions will evolve this 'AI brain', as will agentic AI that allows for greater AI reasoning," he forecast.

"To be sure, don’t expect to immediately see intelligent robots in homes, restaurants, service areas and factories. But these use cases may be closer than you think, as governments look for solutions to ageing societies and shrinking labour pools. Physical automation is going to happen gradually, in 10 years being as ubiquitous as the iPhone."

Source: NVIDIA. Lebaredian.

Rev Lebaredian, VP of Omniverse and Simulation Technology, NVIDIA, pointed out that humanoid robots are not common because of affordability. "As agentic AI brings new intelligence to robots, though, volume will pick up and costs will come down sharply. The average cost of industrial robots is expected to drop to US$10,800 in 2025, down sharply from US$46K in 2010 to US$27K in 2017. As these devices become significantly cheaper, they’ll become as commonplace across industries as mobile devices are," he said.

Explore

This is the 2nd of a three-part series on AI predictions for 2025. Read part 1 and part 3.

More AI-related predictions can be found in posts on agentic AI as well as AI cybersecurity, as well as throughout the 2-Z of 2025 predictions series.

Data minimisation is also mentioned under D is for Data Management.

Hashtag: #2025Predictions