AI agents promise to move enterprise technology beyond storing and processing information towards systems that can understand context, make decisions and take action. But what needs to be in place before organisations can put agents to work on scale?
While the emergence of agentic AI bodes well for enhancing the efficiencies of enterprises across diverse industries, organisations that adopt these platforms can really get a return on their investments if these systems are integrated into the workflow processes and not simply relegated to run alongside legacy systems.
These remarks were made by Jannie Malan, principal solution consultant at Workday South Africa at the recent 2026 IT Indaba held at the Sandton Convention Centre when he presented at a session titled: Ready for agents? Building the foundations for organisational AI.
Agentic AI refers to autonomous software systems that set or interpret goals, make multi-step plans, use external tools, and execute workflows with minimal human oversight. Instead of just responding to a prompt with text, agentic AI actively reasons, acts, evaluates results, and adjusts its path until a goal is achieved.
Agentic AI is used in various applications such as, among others, software development, customer support and operations, supply chain logistics and finance.
Drawing parallels to the industrial revolution when electricity was introduced as a substitute power source for steam engines in the 1890s, Jannie said the introduction of electricity did not result in any noticeable improvement in productivity because the factory managers merely still retained old workflows. “The factory managers simply retrofitted their factories by unhooking steam drives and bolting electric motors onto the same belt shaft architecture. They opted for a cheaper fuel source and not a different factory outlay,” he said.
Putting elements in place
He noted that despite the new power source, productivity in these factories remained stagnant. However, a breakthrough occurred when managers overhauled the old layout and built single storey plants around individual unit drivers.
Like what occurred in the 20th century, the return on investment for agentic AI will materialise the moment it is built onto the workflow and integrated to run on your data, your rules, and your approvals.
He said though the research has found that agentic AI applications can generate savings of up to 80 percent in time, it becomes pointless when the time saved is invested into doing more tasks inside the same processes. “Doing more of the wrong tasks faster is still doing the wrong tasks,” he said.
He advised that there are foundational elements that need to be firmly in place for agentic AI to work on an enterprise-wide scale. “AI’s real value is leveraging over scarce human attention – this isn’t about being marginally faster or cheaper. It’s about removing drag from how work moves,” he added.
Firstly, the application should be trusted by business and prospective users should feel confident to adopt it because of adherence to stringent governance, controls, audit trails and safety. Secondly Jannie said it needs to be actionable and provide access to workflows and processes. “In other words, it needs to be wired into the processes where the work happens and not just sit alongside them,” Jannie said.
Thirdly it must facilitate access and possess the proprietary context that is grounded in your data and business logic so that it produces outcomes that are right and customised for your business. “All these three elements are required for a successful agentic AI deployment. If you miss one, it will stay as a demo, if you have all three it defines how work gets done,” he said.
Human intervention key
While the emergence of agentic AI platforms marks a step in the right direction, Jannie cautioned that some of these off the shelf solutions can open a gateway to activities they were not intended for such as generating unauthorised payments, making illegal scheduling, and being susceptible to sensitive data leakage. “They become a blueprint for chaos,” Jannie warned.
Citing an example of a simulation where an agentic AI application was requested to source a contractor and draft a work contract, Jannie noted that though the application did an excellent job of undertaking this task in line with the prompts, it was “lawless” and skipped the mandatory background checks, ignored the in-country pay scales, and side-stepped approval for pay exception.
He said the Workday solutions always require human intervention and approvals which negates concerns about them going rogue.
He said an added benefit of Workday platform is that it allows software developers to plug into their AI agentic tool of choice and build software using natural language which enables users to generate prompts using the same conversational, natural language.
