Artificial intelligence (AI) is entering a new operational phase as companies begin testing systems designed not just to respond to prompts but to plan tasks, make decisions and execute actions continuously within business environments.

While most organisations have so far adopted AI as a productivity layer, the model is now shifting as agentic AI systems are developed to pursue defined goals over time and coordinate multiple actions rather than stop after completing single tasks.

Alongside this shift, research into artificial general intelligence (AGI) is expanding beyond narrow task optimisation toward systems capable of transferring knowledge across domains, a trajectory that could fundamentally alter how digital work is organised.

Global market intelligence firm The Business Research Company notes that one of the main factors propelling the AGI market is the rising investment in advanced AI research.

“These investments focus on developing AI systems that perform complex tasks, learn from varied data, and exhibit reasoning and problem-solving on par with human cognition,” notes the firm in a report.

“Governments, private companies and academic institutions are investing heavily to build AI that can operate autonomously, improve decision-making, and foster innovation across multiple industries.”

Tech analysts observe that the transition is being driven by practical pressures, as businesses seek faster decision cycles, lower operational costs, and automation that extends beyond execution into planning and coordination.

According to Anthony Muiyuro, East Africa’s regional director at Syntura, the defining change is not intelligence itself but continuity of action within structured environments.

“Today’s AI tools are largely reactive. They respond to prompts, execute narrow tasks and stop. Agentic systems, on the other hand, represent a shift toward proactive, goal-driven AI systems that can plan, make decisions, use tools, learn from outcomes, and act continuously with limited human intervention,” says Mr Muiyuro.

The techie observes that the shift has become commercially viable due to a combination of stronger models, lower computing costs, and rising demand from firms seeking automation that delivers measurable efficiency gains.

Capabilities that signal movement toward autonomous reasoning include multi-step planning, goal prioritisation where systems decide what matters most under constraints, self-correction without retraining, as well as the ability to select and use software tools independently.

US-based Tesla billionaire Elon Musk has previously predicted that AGI will surpass total human intelligence by 2030, describing AI and robotics as a "supersonic tsunami" leading to a "technological singularity," where AI iterates beyond human comprehension.

"Humans are currently in the 'biological bootloader' phase of digital superintelligence… This supersonic tsunami-like change no longer allows us to press the pause button," said Mr Musk in a past interview.

Today, early versions of these capabilities already exist, but deployment remains limited as organisations test reliability and attempt to manage risks associated with autonomy at scale.

Industry specialists note that the constraint is less about technical performance and more about predictability, governance, and accountability when systems are allowed to act without continuous human oversight.

“Expect narrow autonomous agents at scale within two to three years, with broader reasoning systems following more cautiously,” predicts Mr Muiyuro.

As a result, early adoption is currently concentrated in roles that are knowledge-heavy, repeatable, and fully digital, where outcomes can be clearly defined and measured.

These include customer support operations, software maintenance, IT infrastructure management, compliance checks, and structured financial reporting functions, among others.

Such roles provide controlled environments in which objectives are explicit, data flows are logged, and errors can be detected before cascading through wider systems.

In contrast, roles that rely heavily on judgment, trust, or social context remain more insulated from near-term automation.

Services such as healthcare delivery, caregiving, diplomacy, and governance continue to require human discretion that is difficult to encode into autonomous systems. Rather than eliminating jobs, Nr Muiyuro says, agentic AI is expected to recompose roles by shifting routine execution to machines while concentrating human effort on oversight, exception handling, and strategic judgment.

While this restructuring creates productivity gains, it also introduces operational risks that differ from those associated with earlier automation technologies. One foreseeable risk is goal misalignment, where systems optimise defined objectives that do not fully reflect organisational intent or ethical boundaries.

Others include opacity, as autonomous decision paths can become difficult to audit or explain, with error amplification presenting further concerns, as small mistakes could propagate rapidly when systems operate continuously across interconnected processes.

Over-delegation and enhanced security exposure have also been cited as emerging risks, with humans potentially losing situational awareness as responsibility shifts incrementally toward autonomous systems.

Agentic systems often require access to sensitive data, credentials, and internal tools to function effectively, which, if compromised, could provide attackers with deeper access than traditional software applications.

Mr Muiyuro says governance frameworks have not kept pace with technological capability, particularly in emerging markets where regulatory capacity remains uneven.

“While technical safeguards are improving, governance frameworks are lagging - especially in emerging markets. Regulation, board oversight, and accountability models are still catching up to the reality of autonomous systems operating inside businesses,” he says.

“The key challenge is not whether agentic AI will arrive, but whether organisations adopt it deliberately, responsibly, and with human control firmly in the loop.”

Globally, firms deploying agentic AI are doing so cautiously, often limiting decision authority and requiring human approval at critical stages.

Most deployments begin with narrow use cases, parallel human supervision, and extensive logging to evaluate performance and risk exposure.

→ kmwangi@ke.nationmedia.com