Thanks to the increasing accessibility of large language models (LLMs) in recent years, more users are now interacting with AI-powered applications such as ChatGPT. In March, OpenAI released a major upgrade for its GPT-4o model to allow users to natively generate or modify images or photos, claiming that the model is capable of “precise, accurate and photorealistic outputs”.

A 2024 Work Trend Index by Microsoft and LinkedIn also revealed that 84% of people in Malaysia have started using AI to save time at work. It has also led to the growth of ‘AI power users’ or people using AI frequently on a daily basis.

While AI has proven to be ­useful for generating images and helping employees be more productive at work, its potential can extend far beyond content creation and productivity tweaks.

Monash University Malaysia School of Business associate professor Dr Manjeevan Singh Seera says the next evolution for AI is Agentic AI, which is a system capable of making decisions independently based on data and what it has learned, rather than just helping humans make choices.

“Current AI models require a lot of human intervention, where you have to key in inputs (or prompts) and make adjustments along the way as it may not always give the right solutions.

“It mostly provides suggestions or support but agentic AI could actually handle tasks on its own, especially routine ones, without human oversight,” he says in an interview with LifestyleTech.

The next big thing?

A Bloomberg report in 2024 describes “agents that do work for you” as Big Tech’s new AI obsession, with entities like software company Salesforce signing a deal to install AI agents at more than 200 companies including Fedex and Accenture.

The report also shares details about how an insurance company has deployed an AI agent to handle 60% of customer requests – communicating via phone, email or chat – and was trained on 300 hours of human voice.

In a podcast last year, OpenAI CEO Sam Altman says agents are “the next giant breakthrough”, describing the prospect as akin to collaborating with a “really smart senior coworker” that can work together with users on a project.

Manjeevan says a shift towards developing agentic AI can be seen through OpenAI’s introduction of Deep Research in February. The company describes it as an “agent that can do your work for you independently” where through a prompt on ChatGPT, it will find, analyse and synthesise hundreds of online sources to create a ­­comprehensive report at a more detailed level than the usual chat response.

With more access to tools like Deep Research and Manus, Manjeevan says it’s possible for Agentic AI to simplify life by taking over complicated tasks like tax filing. — ASSOC PROF DR MANJEEVAN SINGH SEERA

“For example, say you need to plan a trip to Japan. You key in a prompt with details about your trip and what you have in mind. It will do the research for you and provide a comprehensive travel itinerary,” Manjeevan says.

In March, China-based startup Butterfly Effect released Manus, an agentic AI platform where it could reportedly perform real-world tasks like organise an interview session with 40 candidates, prepare a report to compare insurance policies and more. Some users on X say they have used Manus to perform lengthy research and built a website based on the result it presented.

With more access to tools like Deep Research and Manus, Manjeevan says it’s possible for Agentic AI to eventually simplify life by taking over complicated tasks like tax filing.

“Imagine an AI system automatically analysing your receipts, calculating deductions, preparing tax forms and submitting them. It could help you to save time and reduce mistakes.

“Currently, users have to key in details manually and figure out on their own what is claimable for tax filing. But (I believe) with Agentic AI, you can just provide all your receipts and it can tell what is a business expense, personal and so on,” he adds.

Recent reports support this vision with firms like PWC announcing earlier this year that it has developed AI agents that can gather financial data, analyse tax laws and generate reports with minimal human input.

Similarly, Salesforce says as Americans spend an average 13 hours filing taxes every year, AI agents could reduce that time by 62%.

Double agent?

Is Malaysia ready for Agentic AI? Manjeevan says there is work that needs to be done before Agentic AI can be widely adopted.

“The main challenges are creating clear regulations and earning public trust. Industries like banking, e-commerce, and healthcare would likely gain the most, as these sectors deal with complex tasks where AI can make a big difference,” he says.

Manjeevan says Malaysians will also likely need time to fully trust AI with more complex and sensitive tasks. He also wonders who will be responsible when Agentic AI is misused – for example, when it has the autonomy to generate and publish potentially harmful content?

“There should be a collaborative effort between the public and private sector in establishing public trust by addressing possible concerns. The National AI Office here can take the lead in fostering this collaboration,” he adds.

With Malaysia’s diverse population, Manjeevan also stresses the need for AI systems to be trained on a broad range of data, particularly in local languages and cultural contexts. Since most AI tools are developed in foreign countries, they may struggle to accurately represent certain segments of the population or even exclude them due to gaps in the training data.

“One major risk is AI systems reflecting biases or unfairness that exist in their training data, leading to unfair or inaccurate decisions. It’s essential to extensively test AI systems before launching them,” he says.

For Malaysia to become an AI-driven nation, Manjeevan says transparency is crucial as consumers will be able to trust the technology more when they understand its function, how it makes decisions and be able to see for themselves how it performs reliably with consistency.

“Clear rules, regular independent audits, and openly designed AI systems would help ensure fairness and reduce misuse,” he adds.