AI is being used across workplace systems, software development, manufacturing, customer engagement, and critical infrastructure. Its wider deployment is placing greater attention on how organisations divide responsibilities between employees and AI, secure the underlying data, and maintain oversight of automated decisions.
AI Appreciation Day provides an opportunity to examine these issues as companies move beyond standalone tools and limited trials.
“AI Appreciation Day provides an opportunity to recognise that the true value of artificial intelligence lies not in the technology itself, but in its ability to empower people to focus on more meaningful, strategic, and high-value work,” Jessica Zhang, senior vice president at ADP APAC, said.
ADP’s research found that around half of workers globally use AI several times a week, while one in five use it almost daily. In Singapore, nearly one-quarter of workers use AI almost every day, while more than half engage with it several times a week.
Organisations are now examining individual roles to determine which tasks can be handled by AI and which continue to depend on human capabilities. ADP refers to this approach as “The Great Job Unbundling.”
Routine administrative work can be assigned to automated systems, while employees retain responsibility for critical thinking, collaboration, creativity, and interpreting AI-generated information. Zhang also identified practical training and clear communication about changing roles as requirements for managing that transition.
Software engineering is one area where this division of work is affecting day-to-day responsibilities. Engineers are spending less time writing routine code from scratch and more time directing, reviewing, and validating work completed by AI systems, according to Richard Spence, vice president and general manager for Asia-Pacific at Cognition.
“Coding ability remains fundamental, but judgement, system design, and orchestrating end-to-end software delivery are becoming equally important,” Spence said.
Engineers must decide which assignments can be delegated, review the resulting code, and confirm that it meets security, performance, and reliability requirements. AI can change how code is produced, but accountability for the finished software remains with the engineering team.
That responsibility extends to the software supply chain. AI agents can introduce packages, models, development tools, and suggested fixes into engineering workflows, creating additional assets that organisations must monitor.
JFrog’s 2026 Software Supply Chain Security State of the Union report found that 83% of organisations surveyed in Asia-Pacific had checks for AI inputs and outputs, 12 percentage points above the global average. However, only 55% used automated systems to detect unauthorised or unapproved AI use.
“AI deserves appreciation. But in the enterprise, it also requires evidence,” Yashaswi Mudumbai, senior director of solution engineering for Asia-Pacific at JFrog, said.
Governance records need to cover software packages, AI models, agent skills, and Model Context Protocol servers. Automated enforcement can also help determine whether an AI-generated asset complies with security and development policies before it enters production.
AI agents introduce further control requirements because they can perform tasks across several business systems rather than only generate text, code, or recommendations.
Kyndryl’s 2026 People Readiness Report found that 57% of organisations had deployed AI broadly or embedded it in core business processes. Only 32% had achieved at least one of their two main AI objectives.
The report also found that 81% expected AI agents to make consequential decisions within the following year. However, only 25% completely trusted AI systems operating without human oversight.
“The organisations pulling ahead will be those that treat AI as an operating model change, not a technology rollout,” Dr Vishnu Nanduri, AI innovation leader for ASEAN and Korea at Kyndryl, said.
Defined roles, permissions, escalation procedures, and audit records are needed before agents are allowed to act across IT, finance, operations, supply chains, and customer service. Businesses also need to determine how those systems interact with employees, applications, and approval processes.
Agents require access to current business information if they are expected to make decisions or trigger actions. Models operating across isolated applications and databases can lack the context needed to coordinate work.
“AI will not create business value because organisations deploy more models — its value amplifies when those models can act on trusted, real-time context,” Greg Taylor, senior vice president for Asia-Pacific at Confluent, said.
Confluent’s Data Streaming Report found that 75% of IT leaders surveyed in Singapore were deploying or piloting agentic AI. Another 78% said insufficient real-time data infrastructure was slowing their AI programmes.
Organisations therefore need to connect information across business functions, control how it is accessed, and keep it updated. Without that foundation, agents can produce outputs based on incomplete information or remain unable to work with other systems.
Operational context is also necessary when AI is incorporated into supply-chain, production, financial, or enterprise resource planning workflows. These systems need to interpret company-specific data, process rules, and approval requirements rather than operate as general-purpose assistants.
“The organisations seeing the greatest value from AI are those embedding it into the operational workflows that power their operations, rather than treating it as a standalone productivity tool,” Geoff Thomas, senior vice president and general manager for Asia-Pacific and Japan at Infor, said.
Thomas cited production continuity requirements at Japanese automotive manufacturers, supply-chain changes affecting Malaysian semiconductor producers, and traceability obligations for Australian food companies.
These applications require AI systems to operate with sector-specific processes and data. Business rules and operational records provide the context needed for AI to support defined actions rather than produce general recommendations.
Industrial AI carries additional requirements because its outputs can affect machinery, production schedules, product quality, and worker safety.
Singapore’s Budget 2026 introduced national AI missions covering advanced manufacturing, connectivity, finance, and healthcare. The manufacturing mission includes plans to apply AI across factory and engineering environments and support wider adoption within the sector.
“The challenge is not simply adopting AI, but embedding trusted AI into the workflows that design, simulate, build, and operate products,” Alex Teo, vice president and managing director for Southeast Asia at Siemens Digital Industries Software, said.
Industrial deployments require trusted production data, transparent outputs, and governance processes that allow engineers to validate AI-generated recommendations. Human responsibility remains necessary when those recommendations are used to adjust equipment, alter designs, or change production decisions.
The requirements are more stringent in aviation, defence, public safety, and critical infrastructure, where incorrect or delayed decisions can affect physical systems and essential services.
“The real test is whether AI can operate safely, securely, and reliably when decisions carry real-world consequences,” Emily Tan, CEO of Thales Solutions Asia and country director for Thales in Singapore, said.
AI cannot be treated as a separate software component in these environments. It must be engineered alongside the sensors, data systems, communications networks, cybersecurity controls, and personnel supporting the wider operation.
Its outputs must also be explainable and verifiable under operational conditions. Critical systems are generally built around resilience, validation, redundancy, traceability, and human oversight.
Customer-facing AI introduces a different trust issue. The ability to generate more marketing content does not mean customers will regard the resulting interactions as accurate, relevant, or useful.
“AI has become remarkably good at generating content. It’s still much harder to generate trust,” Shahid Nizami, vice president for Asia-Pacific and the Gulf Cooperation Council at Braze, said.
Braze’s 2026 Customer Engagement Review found that 93% of surveyed marketers believed AI helped them understand customers better. Only 53% of customers said brands accurately predicted their wants and needs.
The difference places greater attention on how companies use customer data, behavioural signals, timing, and message frequency. Human judgement remains necessary to prevent AI-supported engagement from becoming intrusive, repetitive, or disconnected from customer expectations.
Production AI also depends on the infrastructure used to process, move, store, and secure data. Computing capacity and network connectivity must be accompanied by controls governing where corporate information and AI-generated knowledge are retained.
“The shift from training AI models to running AI in production is changing what organisations expect from their infrastructure,” Govind Choudhary, general manager for Southeast Asia and India at Digital Realty, said.
Businesses are looking beyond where data is stored to consider how knowledge produced by AI is governed and protected. Infrastructure supporting production AI must provide performance and connectivity while keeping information within the organisation’s security and compliance boundaries.
Singapore’s proposed Digital Infrastructure Bill covers major data-centre operators and cloud-service providers. The draft framework includes requirements relating to cybersecurity, physical security, business continuity, disaster recovery, incident reporting, and environmental performance.
Governance must also remain consistent when AI workloads are distributed across private infrastructure, public clouds, and edge environments.
“Organisations need consistent visibility into how AI systems are developed, deployed, and managed, regardless of where workloads run,” Juliana Lim, country manager at Red Hat Singapore, said.
This requires organisations to track where AI systems operate, how they are updated, which data they access, and whether autonomous actions remain auditable under internal and regulatory requirements.
Want to experience the full spectrum of enterprise technology innovation? Join TechEx in Amsterdam, California, and London. Covering AI, Big Data, Cyber Security, IoT, Digital Transformation, Intelligent Automation, Edge Computing, and Data Centres, TechEx brings together global leaders to share real-world use cases and in-depth insights. Click here for more information.
Tech Wire Asia is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。