AI agents, which are specially designed AI software tools, will automate much of the routine tasks people do daily and will also completely take over certain routine tasks and processes, said Microsoft South Africa AI modern work use senior architect Michael Mullany.

Agentic AI is the next evolution of applied AI because it not only assists the user but can act on his or her behalf.

Research and consulting firm Gartner said that, by 2027, 33% of enterprise software applications would include agentic AI and that 15% of daily work would be done autonomously by agentic AI solutions, he added during a briefing on June 4.

Highlighting the rapid adoption of these solutions, research conducted by Microsoft has found that 82% of business leaders are confident they will use AI agents in digital teams as part of their workforce in the next 12 to 18 months, Mullany noted.

Specifically, these AI agents are similar to Microsoft's AI powered personal assistance tool Copilot, but are tailored to specific business use cases and processes.

They are more honed, targeted and directed than general use AI tools and are focused on the data applicable to the business task or tasks they are assigned to, he explained.

AI agents have specialised skills or tasks that they can perform, such as sending confirmation emails, updating records or automatically updating digital devices, among others, he illustrated.

These tools use the same data that the employee has permission to access. These agents also inherit the employee's cybersecurity permissions and can seamlessly integrate into companies' digital environments, he said.

“For example, an airport would integrate airline data into its AI agent solutions for staff, while a bank would integrate transactional and financial data into its systems, among others.”

The AI agents fulfil a range of roles, from basic routine work such as updating records to advanced AI agents that can produce researched documents that can be used in enterprises.

During the briefing to journalists, Mullany demonstrated the use of four AI agents. The first was the idea coach AI agent that he used to recommend various means of addressing any additional IT-demand in the manufacturing and mining sectors that the company had not yet tapped into.

The second he demonstrated was the writing coach AI agent that helped him to summarise and set out in steps how a business could register to work with government, using verbal instructions he had received.

This verbal voice note was converted into text that the agent then used alongside public information available on government websites to create the step-by-step instructions from the note, he explained.

The third AI agent he demonstrated was the analyst agent.

He provided it with a spreadsheet containing a range of customer information. He then asked the agent to produce a report from this vast amount of information and received a breakdown of the return on investment per customer segment and according to age, and the amount of profit each segment generated, among others.

Further, he demonstrated an advanced AI agent, namely the research agent. He asked it to research frontier firms, which is a topic Microsoft CEO and chairperson Satya Nadella has spoken about.

Based on deep research models and Open AI models and using the Microsoft data that Mullany has access to, as well as publicly available data and opinion pieces, the output from the agent presented first a definition of what a frontier firm is, as well as the five key structures. It also presented the list of sources it used.

“AI agents will change the functions and processes across organisations. [Information and communications technology market research firm] the International Data Corporation estimates that one-billion new AI agents will be built over the next four years.

“AI is fast becoming a business necessity and can automate and unlock new value in organisations,” said Mullany.

He highlighted that Microsoft's legal department had built its own AI agent to handle routine queries, such as discounts when customers pay in a different currency, which are important albeit routine.

“These topics are often complex or uncertain for those in other roles; but not so complicated for the lawyers. They prefer spending their time on legal and regulatory questions and work for the company, and have the AI agent handle the more common queries.”

Meanwhile, companies can use Copilot studios to build their own AI agents for their own uses.

“All Microsoft's AI solutions adhere to our responsible AI framework. This means the solutions have built-in guardrails and safety features aligned to ethical and responsible AI uses,” he said.

Another feature of Microsoft's responsible AI is transparency, with administrators having full view of how the data is handled and processed. He pointed out that these safeguards are necessary for uses such as healthcare to ensure that all patient data are handled responsibly and that the outputs are safe, secure and compliant.