AI shifts to governed, real-time systems at the edge

Technology vendors and automation specialists are using AI Appreciation Day to highlight a shift from experimental deployments to large-scale, tightly governed AI systems. Their comments point to a shared focus on reliability, transparency and control as artificial intelligence becomes embedded in critical processes and consumer services.

In the enterprise, AI is moving from the data centre into production environments at the network edge. There, models increasingly observe physical conditions and trigger actions in real time. This shift is prompting new governance frameworks as organisations consider how to monitor autonomous decision-making and mitigate risk across hybrid IT estates.

Fumiki Negishi, Vice President & General Manager, HPC & AI APJ GTM Division at HPE, describes this as a move into an "inference-driven era" in which physicality and locality define deployment models. AI systems no longer just analyse data. They interact with the physical world, respond to sensor input and operate across distributed infrastructure spanning multiple jurisdictions and organisational boundaries.

"AI is entering an inference-driven era, extending beyond the data centre to the edge where decisions are made. This shift is defined by two dimensions: physicality and locality. Physicality reflects AI's evolution from generating insights to understanding and responding to the physical world through multimodal awareness and real-time feedback, making low latency, governance, and ethical safeguards essential. Locality comes as the edge is ubiquitous and the sovereignty of AI will evolve into multiple layers of society, culture, identity, and ultimately personality, of AI interwoven with the human side of this structure. To keep pace, enterprises across Asia Pacific are adopting new operating models that enable AI to scale responsibly and deliver real-world impact," said Fumiki Negishi, Vice President & General Manager, HPC & AI APJ GTM Division, HPE.

He argues that the issue is no longer proof of concept but scale. Enterprises across Asia-Pacific are moving from pilots to production systems that can act autonomously across plants, branches and digital channels. That change puts governance on the critical path.

Negishi points to research suggesting that few organisations have reached maturity in this area.

"Today, systems are no longer designed only to process information, but are increasingly expected to observe, decide, and act in real time. This ability to operate with intelligence and autonomy is what gives the agentic enterprise its advantage: responding faster, adapting earlier, and operating at a scale no human team alone could sustain. The challenge now is not proving AI can work. It is scaling from isolated pilots to production-grade systems that can make real decisions on a real scale, without requiring human oversight at every step. It is governance that turns intelligence into something enterprises can trust with real decisions. In fact, as per Deloitte's report - State of AI in the Enterprise, (https://www.deloitte.com/content/dam/assets-zone3/us/en/docs/services/consulting/2026/state-of-ai-2026.pdf) just 21% of organisations report having a mature governance framework for autonomous agents, even as adoption accelerates across the enterprise. That imbalance highlights one of the defining challenges of the agentic AI era: scaling intelligence faster than organisations can govern it. As AI systems move from pilots into production environments, robust governance will increasingly determine whether enterprises can capture value at scale while maintaining trust, accountability, and control. Governance is not a brake on AI ambition, but the foundation that allows intelligent systems to operate with speed, autonomy, and trust once decisions are being made independently at production scale. That discipline is what makes agentic AI scalable and sustainable across the enterprise. AI factories represent the backbone of the agentic enterprise, where data pipelines, model training, and inference come together to manufacture intelligence at scale. Here, agentic governance means the ability to monitor what an agent does once it's live, enforce policy in real time, and recover cleanly if something goes wrong. That operational maturity is what turns AI from a promising pilot into a system enterprises can genuinely depend on. In hybrid IT operations, agentic AIOps is unifying what used to be a sprawl of disconnected agents into a single, coordinated, observable control plane spanning cloud and on-premises environments alike. Enterprises now have greater visibility into agent actions and clearer accountability across systems. Increasingly, that discipline is extending to the network itself, where self-driving infrastructure can analyse telemetry, detect anomalies, and remediate issues autonomously under policy and at a speed no manual process could match. Asia-Pacific's agentic AI journey is accelerating, and that's worth marking this AI Appreciation Day. The organizations setting the pace aren't just the ones moving fastest. They're the ones building the governance to keep scaling, turning intelligence into something the enterprise can depend on, not just deploy," said Negishi.

Process design and data quality sit alongside governance as foundational concerns. Automation specialist Nintex warns that organisations rushing into multi-agent architectures risk magnifying weaknesses in how work flows across departments.

Chris Ellis, Director Solution Engineering at Nintex, stresses that agent networks inherit the strengths and flaws of the processes they execute.

"As organisations shift from initial AI experiments to multi-agent networks at scale, AI Appreciation Day is a chance to consider the foundations that will make this transformation a reality. Multi-agent systems accelerate execution, but they do not automatically improve the underlying process. If an organisation has inconsistent procedures or conflicting policies, those weaknesses will be amplified by AI agents operating at scale. Before deploying agent networks, organisations must first understand, document, and optimise the underlying processes. The principle is the same as traditional automation projects: garbage in means garbage out. Agents can only make decisions based on the information, rules, and context they are provided. Poor data quality or undocumented business rules can quickly propagate across an entire agent network, creating errors at a pace impossible in manual environments. When something goes wrong, you should be able to reconstruct the decision path rather than treat the system as a black box. This matters as much for governance as it does for debugging. More importantly, organisations can use these insights to continuously improve both the process and the agent network itself. Multi-agent AI is the near-term future of enterprise automation. To prepare for this future, organisations must think about how to deploy it in a way that captures the rewards while avoiding the risks. At the end of the day, trustworthy AI outcomes are built on trustworthy processes," said Chris Ellis, Director Solution Engineering, Nintex.

Customer engagement platforms are positioning AI as a front-line communications tool, moving beyond scripted chatbots. Brands are increasingly integrating conversational interfaces and voice agents into contact channels as they pursue personalisation without higher labour costs.

Damien Brennan, Strategic AI and Emerging Tech Partnership Manager - APAC at Sinch, says this marks a departure from earlier automation-first narratives.

"AI has become a powerful customer communications tool, that is enabling brands to build stronger connections with their customers, rather than simply managing a higher volume of requests. This marks a significant evolution beyond its early promise of being an 'automation at scale' tool. Developments in conversational AI mean customers can now reach a brand at many more times in the day, with responses that understand context and intent rather than matching keywords to scripts. That same understanding extends naturally into voice AI, which now supports real-time conversation, allowing customers to speak with a brand without navigating rigid menu structures. Collectively, these capabilities allow brands to remain consistently present for customers, without compromising the importance of personalisation in interactions. The results are measurable, including faster resolution times and stronger engagement, and these outcomes are precisely why AI has become embedded in how organisations communicate. AI Appreciation Day offers a timely opportunity to recognise that progress. Conversational and voice AI are already strengthening customer connection at scale, underscoring how far the technology has advanced beyond simple automation," said Damien Brennan, Strategic AI and Emerging Tech Partnership Manager - APAC, Sinch.

In the consumer market, vendors describe a similar shift in expectations around security products. Buyers are now questioning long-term reliability, data handling and subscription models, not just headline features.

Home security specialist Reolink sees that change in demand for on-device analytics and subscription-free services.

"AI Appreciation Day 2026 arrives at a genuinely interesting moment for the home security industry. A year ago, the conversation centred on what AI could do: object detection, false alarm reduction, and smart alerts. The more important question in 2026 is whether those capabilities are still working for them. Consumer expectations have matured after years of cloud outages and subscription fatigue. People want systems that understand context, surface what genuinely matters, and provide meaningful alerts for important moments. The industry's broader shift toward on-device processing reflects where those expectations are heading. Processing footage locally means faster responses and stronger protection against the vulnerabilities that come with storing data in the cloud. Reolink's own Local AI Video Search reflects this thinking, running entirely on the device so there's no footage sent elsewhere and no data sitting on external servers exposed to breach or interception. Keeping these AI features subscription-free is also a deliberate commitment. As AI becomes more widely available, we know our consumers will want capable AI without an ongoing payment plan. The category is moving toward something people can genuinely rely on: capable, private by design, and accessible regardless of the hardware they already own," said Nick Nigro, Vice President Sales Australasia, Reolink.