From discovery to delivery: what agentic commerce changes first

Agentic commerce is often described as “shopping through AI.” That’s true, but incomplete.

What’s changing is not just where customers search. It’s the entire chain: how products get discovered, how decisions are made, how payments clear, and how orders are fulfilled and supported, often with AI agents handling parts of the workflow behind the scenes.

Not every layer is changing in the same way. Discovery is the most visible shift. Checkout is evolving structurally. Operations and logistics are already seeing practical automation.

The companies moving fastest are not waiting for full autonomy. They are deploying agentic workflows wherever they can create leverage today.

Discovery becomes curation

The first shift most teams notice is discovery.

Traditional ecommerce was built for browsing. Category pages. Filters. Endless scroll.

Agentic discovery collapses that loop. Customers show up with intent. The system narrows the universe quickly.

As Amit Sagiv, Co-Head of Payments at Wix, pointed out, “When you go to ChatGPT and you’re looking for something, you… get a very granular ask.”

Granular asks require structured answers. In this environment, discovery becomes a data problem before it becomes a marketing problem. The system is matching intent to attributes. 

Jake Sinsheimer, Head of Agentic Commerce, GTM at Stripe, put it directly, “Agents need more structured data, and most product data tends to be what I would refer to as semi structured or just optimized for human reading.”

Most ecommerce catalogs were designed for humans skimming pages. Agentic systems require clean inputs.

Geoff Smalling, Chief Technology Officer at Shippo, sees the downstream impact. He described how incorrect product dimensions or inaccurate attributes create friction across ecommerce sites. In a system that explicitly filters on width, depth, stock availability, and delivery timing, bad data does not just look sloppy. It gets excluded.

The implication is clear:

-Catalog hygiene becomes a growth lever.
-Attribute completeness becomes distribution.
-Structured data becomes a competitive advantage.

Discovery is no longer about who has the best homepage. It is about who has the cleanest inputs.

Checkout becomes the moment trust is enforced

As discovery compresses, commitment becomes more consequential.

Checkout is no longer just a page with fields. It becomes the control system for the transaction.

At this layer, the system evaluates:

• Payment credentials
  
  
• Fraud and risk signals
  
  
• Inventory availability
  
  
• Delivery promises
  
  
• Budget guardrails
  

Trust builds incrementally. Lower-risk purchases with clear constraints establish confidence. Over time, higher-stakes transactions can move with less friction.

Amit captured the design tension clearly, “On the one side, you want to eliminate the checkout because it’s a friction point. But you want to have very clear control over the action.”

Geoff added the context layer that makes this powerful. For example, a high-value gift with a tight delivery window should trigger different shipping and insurance recommendations than a routine reorder. When checkout incorporates context, it improves decision quality rather than simply speeding up the transaction.

For operators, this reframes the goal. It is not about removing steps. It is about defining guardrails:

• When should a transaction auto-approve?
  
  
• When should a human review be required?
  
  
• When should delivery upgrades surface automatically?
  
  
• Where should risk thresholds block fulfillment?
  

As interfaces get smarter, enforcement layers need to be disciplined.

Operations and logistics are already seeing leverage

While headlines focus on AI-powered shopping experiences, the most immediate returns often show up inside the business.

Geoff expected early experimentation to center on customer-facing workflows like returns. Instead, he is seeing operators focus on internal acceleration.

Support triage. Refund automation. Faster access to order data. Shipping optimization. Identifying high-cost ticket categories and reducing them.

These workflows share three characteristics: they are repetitive, measurable, and already defined.

Agentic systems thrive in that environment.

At Shippo, this includes exploring ways to prefill support tickets with contextual data before a human reviews them, automate refunds within defined guardrails, and analyze shipping performance to optimize decisions more quickly.

The pattern is consistent. Start where the workflow is known and the feedback loop is tight. Operations and logistics offer both.

What to do in the next 90 days

You do not need to rebuild your stack. You need readiness.

Jake’s guidance for teams moving quickly comes down to four practical moves:

1. Treat your catalog like infrastructure.
Audit structure, attributes, dimensions, and inventory accuracy. If a system cannot parse your product cleanly, it cannot recommend it.

2. Choose where to test deliberately.
Pick a surface or integration to experiment with. Learn in one environment before expanding.

3. Run a focused pilot.
Select a category or SKU group. Define measurable success. Iterate quickly.

4. Assign ownership.
The companies moving fastest have a clear operator or small team accountable for agentic initiatives. It is not an abstract mandate.

Amit adds a practical filter: start where your team is strongest.

If you are strong in marketing, automate campaign drafts, landing page variations, product description updates, and A/B test variants so your team can test ideas faster. If you are strong in operations, automate repetitive workflows within defined thresholds while keeping a human in the loop for high-risk decisions. If you are strong in merchandising, focus on catalog enrichment, attribute accuracy, and structured data hygiene.

Build internal confidence in controlled environments. Expand from there.