Agentic commerce is coming for your checkout

AI agents won’t just recommend products — they’ll buy them.

Article link: https://open.substack.com/pub/johanosteyn/p/agentic-commerce-is-coming-for-your?r=73gqa&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Audio summary: https://youtu.be/mwqi8KeqSaI

Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/

For most of the internet era, shopping has followed a familiar pattern: search, browse, compare, add to cart, check out. AI has already changed the top of that funnel by helping people discover products and make choices. But the next shift is more consequential. We are moving from AI that advises to AI that acts. In practical terms, that means an AI agent can build your basket, apply discounts, choose delivery options, and place the order.

The experience becomes “tell me what you want” rather than “go through the steps”. Convenience will be the headline. The real story is what happens to control, accountability, and trust when a machine is authorised to spend money on your behalf.

CONTEXT AND BACKGROUND

E-commerce has always been an optimisation game. Retailers obsess over conversion rates, abandoned carts, and friction. Payment providers focus on trust and fraud reduction. Platforms compete for default placement in the customer journey.

Agentic commerce compresses the journey. Instead of sending you to ten websites, an agent can summarise options, make a recommendation, and push a purchase through. That is not just a UI upgrade. It changes market power. Whoever owns the agent interface can become the new gatekeeper of demand, steering customers towards certain merchants, brands, and offers.

It also changes integration requirements. An agent can only transact if it can talk to inventory, pricing, shipping, and payment systems reliably. That is why you are seeing a rush towards standardised “commerce protocols” and tighter partnerships between AI platforms, retailers, and payment providers. The plumbing matters because it determines who can participate and who gets excluded.

INSIGHT AND ANALYSIS

The first tension is consent. Today, you click to confirm. In an agentic world, you delegate. Delegation is not the same as consent, especially when purchases can happen quickly across multiple merchants. We will need clearer controls: spending limits, category restrictions, approval thresholds, and explicit “ask before you buy” rules. Otherwise, convenience becomes a pathway to regret.

The second tension is identity and authentication. If an agent can act as you, identity becomes the core security problem. What counts as “you” authorising a purchase? A password? A device? A voice command? Multi-factor authentication? The wrong answer invites fraud at scale, because an agent is an efficient spender, and criminals love efficiency.

The third tension is liability. If an agent buys the wrong item, who is responsible? The user who delegated? The merchant who fulfilled? The platform that recommended? The agent provider that executed? In traditional e-commerce, accountability is clearer because the user is the actor. In agentic commerce, the actor is a system operating under delegated authority. That will force new contract terms, new dispute processes, and clearer audit trails.

The fourth tension is market fairness. Agents will rank options, and ranking is power. If a platform gets paid to prioritise certain merchants, or if it favours its own ecosystem, we will see a new form of platform bias. The competition question shifts from “who ranks first on search” to “who gets chosen by default when the agent buys”.

Finally, there is a human behaviour issue. Many people already struggle with impulse buying. An agent that reduces friction could amplify that. The checkout page has always been a speed bump where people reconsider. If the speed bump disappears, we will need new consumer protections and better personal controls.

IMPLICATIONS

For consumers, the practical move is to treat agent permissions like financial permissions, not app permissions. Start with strict limits. Require confirmation for every purchase until you genuinely trust the system. Separate “browsing and recommendations” from “ability to transact”. Convenience should be earned.

For retailers, the question is strategic: will you be visible to agents, and on what terms? If agents become a major discovery and purchase channel, merchants will need to integrate, provide accurate product data, and defend their brand experience inside an AI interface that is not fully under their control.

For regulators and policymakers, the priority is clarity: disclosure rules for sponsored recommendations, standards for consent and authentication, and accountability frameworks for disputes. The goal should be to enable innovation while preventing a fraud and manipulation wave.

For business leaders, especially in payments, banking, and customer platforms, agentic commerce should be treated as both an opportunity and a risk. The winners will build trust mechanisms: auditable transactions, clear user controls, and secure identity.

CLOSING TAKEAWAY

Agentic commerce will feel magical because it removes steps. But those steps were not only friction; they were control points. As AI agents move from recommending to buying, the checkout becomes a trust problem, an identity problem, and a governance problem. The organisations that succeed will not be the ones that make purchasing faster at any cost. They will be the ones who make it safe, transparent, and accountable. In the end, the future of commerce will be decided less by how clever the agent is and more by whether people feel comfortable handing it the keys to their wallet.

Author Bio: Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net