Supply Chain, Consumer and Manufacturing in the Age of Agency

Executive summary

• Agents are moving from pilots into operations. Gartner found that supply chain leaders whose companies beat expectations were using significantly more agentic AI across procurement, production, logistics and planning ^[1]. Google, Walmart, Target, Shopify and others have published an open standard so AI agents can complete purchases ^[2]. China expects to produce more than 100,000 humanoid robots this year ^[3].
• Value is shifting towards whoever controls the agent. McKinsey estimates that agentic commerce could orchestrate US$3–5 trillion of global goods sales by 2030 ^[4]. That money will flow through agents that compare prices, switch brands and skip advertising. Manufacturers and retailers that are hard for agents to read or trust will lose share. Brand loyalty will not protect them.
• Governance is the limiting factor, not the technology. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, mainly because of unclear value and weak risk controls ^[5]. Rules are tightening in all three regions. In the EU, AI is now treated as a product for liability purposes ^[6]. Singapore has issued guidance written specifically for agents ^[7]. In the US, courts are testing who may authorise an agent to act ^[8].

What's changing

Autonomous planning and procurement. Gartner surveyed 509 supply chain leaders in late 2025. They named AI-driven changes to ways of working as the most important influence on supply chain strategy over the next two years ^[1]. In June 2026 the World Economic Forum added 16 sites to its Global Lighthouse Network, which recognises leading factories and supply chains. Several now run AI in their core decisions. RRS in Qingdao uses an AI platform for order fulfilment, vehicle scheduling and carrier bidding, and cut transport costs by 23%. Unilever's Haridwar site in India reduced response times by 72% using AI-enabled planning and sourcing ^[9]. Walmart has announced four AI "super agents" covering shoppers, employees, suppliers and developers, built on top of existing tools for dispatch, defect detection and predictive maintenance ^[10].

Physical AI on the factory floor. Robots are becoming both cheaper and more capable. The International Federation of Robotics recorded 542,000 industrial robot installations in 2024, more than double the figure ten years earlier. China accounted for 54% of them, followed by Japan, the US, Korea and Germany ^[11]. Hyundai and Boston Dynamics plan to trial the Atlas humanoid robot on parts sequencing at Hyundai's plant in Georgia, using Google DeepMind's robotics models ^[12]. China's industry ministry says national humanoid output is on track to exceed 100,000 units in 2026. It credits faster progress in large models, agents and chips ^[3].

Agentic commerce. Google launched the Universal Commerce Protocol in January 2026, developed with Walmart, Wayfair, Shopify, Etsy and Target and endorsed by Visa, Mastercard, Stripe, Flipkart and Zalando. It allows shoppers to buy directly inside Google's AI services, with the retailer remaining the seller of record ^[2]. Visa is extending its Intelligent Commerce services across Asia-Pacific with partners including Ant International, Tencent and Microsoft. It has also introduced a protocol that lets merchants tell legitimate agents apart from malicious bots ^[13]. Consumer demand is running ahead of trust. In Checkout.com's research, a third of consumers expect at least 10% of their purchases to be AI-driven within a year. Yet merchants report that agents are involved in only 3% of transactions today ^[14].

Through the Age of Agency lens

Age of Agency: decide who delegates what, and to whom. The board question has changed. It used to be which AI tools to buy. It is now which decisions to hand over to agents, within what limits, and who answers when an agent gets something wrong. Supply chains are well suited to this because so many decisions are high-volume, rules-based and time-sensitive: replenishment, carrier selection, supplier reallocation and production scheduling. The operating model therefore needs written delegation policies, much like financial authority limits, setting out spending caps, escalation triggers and override rights for each class of agent. Consumers want the same controls. In Checkout.com's research, spending caps and instant revocation were the top conditions for trusting a shopping agent ^[14].

Agentrification: watch where value and roles move. There are three shifts to track:

• Margins move to whoever controls the agent. When an agent buys on a consumer's behalf, the brand's relationship with that consumer gets weaker. Some 57% of consumers would let an agent switch brands for better value ^[14]. McKinsey warns that agent-led shopping could bypass retail media networks entirely ^[4]. Platforms that own the agent gain. Brands that depend on advertising, packaging and shelf position lose leverage unless their product data, pricing and fulfilment are clear enough for agents to compare and trust.
• Entry-level roles start to disappear. Some 55% of supply chain leaders expect agentic AI to reduce entry-level hiring, and 86% say they will need new ways to build their talent pipelines ^[1]. Analyst, buyer and planner roles are where future managers learn the business. Cutting them saves money now but leaves a skills gap later.
• Control of the platform matters. The Amazon v. Perplexity injunction shows that a platform can stop a third-party agent from acting for a customer, even when that customer has given permission ^[8]. Firms need to decide whether to build their own agents, take part in open protocols, or both.

UBInt: agent-readiness as essential infrastructure. For the many small and mid-sized manufacturers and distributors across Southeast Asia, Australia and New Zealand, the main risk is not misusing AI. It is becoming invisible to the agent-run buying channels that large buyers and consumers are starting to use. Shared rails such as open commerce protocols, agent-verified payments and machine-readable product data increasingly decide who can take part in trade. Industry bodies and governments can treat access to these rails as basic economic infrastructure. Firms should treat it as a prerequisite for staying in the market, not as an innovation project.

Risks and governance

• Cascading errors. Connected agents can turn one error, such as a wrong order or excess inventory, into failures across the chain ^[4]. Supply chains need limits on how much any one agent can do and must be able to fail safely.
• Liability is moving to the producer. EU member states must transpose the revised Product Liability Directive by 9 December 2026. It treats software and AI as products and AI providers as manufacturers. It also lets courts presume a defect in complex systems ^[6]. Records that show compliance will be the main defence.
• EU AI Act timing has changed. The Digital Omnibus on AI, given final Council approval on 29 June 2026, moves high-risk obligations to December 2027 and, for AI built into regulated products, to August 2028. It also moves AI-enabled machinery to a sector-specific route under the Machinery Regulation ^[15]. The deadlines are later, but the obligations remain.
• Agent authorisation and fraud. Payment and fraud systems were built for human shoppers. They now need ways to verify agents ^[4][13]. In the US, the Amazon v. Perplexity ruling relied on computer-fraud law, which shows that the legal risk of agents acting on platforms is real ^[8].
• Asia-Pacific guidance. Singapore's IMDA published the Model AI Governance Framework for Agentic AI in January 2026. It recommends setting risk boundaries before deployment, least-privilege access, clear human accountability, and step-by-step roll-outs with monitoring ^[7]. Australia's National AI Plan of December 2025 relies on existing law, including consumer protection, privacy and sector-specific rules, together with a new AI Safety Institute ^[16].
• Inflated vendor claims. Gartner estimates that only about 130 of the thousands of vendors claiming to offer agentic AI actually do. Many rebrand chatbots or robotic process automation (RPA) instead ^[5]. Check vendor claims carefully before buying.

Questions for boards and leadership

1. Which operational decisions have we delegated, or are we about to delegate, to agents? Where are the written authority limits, spending caps and override rights for each?
2. If a customer's AI agent compared us with competitors tomorrow, could it read our product data, pricing and availability, and would it trust them?
3. How much of our margin depends on advertising, shelf position or brand preference that agent-led buying could bypass?
4. If we cut entry-level analyst, buyer and planner roles, how will we develop the managers we need in five years?
5. Can we produce the documentation needed to rebut a presumption of defect under the EU's revised liability regime, for our own AI and for our suppliers'?
6. Are we building our own agents, joining open protocols, or depending on platforms that could shut us out?
7. Which of our current agentic AI projects would survive a hard test of value and risk controls, and which should we stop now?

How KIG helps

• KIG Advisory. We help boards and executive teams decide which decisions to delegate to agents and how to govern them. That covers delegation policies, operating-model redesign, agent-readiness for commerce channels, and transformation plans across Asia-Pacific, drawing on collaborators in the US and Europe.
• KIG Research. We develop and apply the Age of Agency, Agentrification and UBInt frameworks. Our research tracks how value, roles and regulation are shifting across manufacturing, logistics and consumer markets, so leaders can act on evidence.
• KIG Ventures. We learn by building and backing AI companies, including autonomous enterprise agents. That practical experience of what works, what fails and what governance agents need in production feeds directly into our advice.