Archive · 4 Sep 2026
Agentic Commerce Needs Human Control As AI Agents Shop
Agentic commerce means an AI helper can shop for someone after receiving permission.
It might find groceries, compare flights or place an order without asking the person every time.
The person still needs to decide what the AI is allowed to do.
For example, they could allow grocery purchases up to ₹2,000 per order and ₹5,000 per month.
The AI should not be allowed to spend more or buy things outside those rules.
People should be able to see what the AI did, pause its permission or cancel it.
Payment companies also need to know which agent is acting and whose permission it has.
In India, these systems are being developed inside merchant apps, messaging services, voice interfaces and third-party assistants.
The goal is to make shopping easier without removing human control.
Agentic commerce lets AI agents discover products, manage carts and place orders under pre-authorized instructions.
India is developing both merchant-led assistants and third-party agents that work across multiple brands.
Controlled delegation would limit agents by spending amount, merchant, category, validity period and other conditions.
Payments systems must identify agents, verify authorization, support multiple payment methods and keep transactions traceable.
Commerce and payment protocols are moving from pilots toward live transactions while emphasizing consumer control and merchant choice.
- Who
- Consumers, merchants, AI agents, payment providers and technology companies are involved; examples include Swiggy, Zepto, MakeMyTrip, ixigo, Juspay, Visa, Mastercard, Google, Stripe and OpenAI.
- What
- The article examines agentic commerce, in which AI agents carry out shopping and payment tasks under controlled human pre-authorization.
- When
- The shift is developing now, with the ecosystem moving from demonstrations and pilots toward transactions involving real buyers, merchants and payment systems.
- Where
- The article focuses significantly on India and also describes a B2B pilot in Singapore.
- Why
- The aim is to make commerce more efficient while ensuring that agents act within user-defined permissions and that transactions remain secure, controllable and auditable.
- India's agentic-commerce models
- Merchant-led assistants operate inside platforms, while third-party assistants work across brands and merchants.
- Example spending instruction
- A consumer could authorize grocery payments up to ₹2,000 per order and ₹5,000 in total per month.
- Required capabilities
- Accuracy, authentication, consumer control and auditability.
- Consumer controls
- Permissions may include spending limits, merchant or category restrictions and validity periods, and should be inspectable, pauseable and revocable.
- Commerce protocols
- Google's Universal Commerce Protocol and the Agentic Commerce Protocol developed by Stripe and OpenAI define ways for agents and merchants to exchange information.
- Payment frameworks
- Visa's Trusted Agent Protocol and Mastercard Agent Pay address agent identity, authorization, tokenized credentials and transaction traceability.
- Singapore pilot
- HSBC and Mastercard completed a B2B pilot connecting corporate buyer, procurement platform SourceSage and supplier FortyTwo, using Mastercard Agent Pay and Juspay technology.