AI Agents vs. The Open Web: Why Agentic Commerce Is Stalling

The Wall Going Up Around Agentic Commerce

Agentic commerce was supposed to be inevitable. AI agents would browse, compare, negotiate, and purchase on behalf of users — collapsing the traditional shopping funnel into a single automated decision loop. Then the internet pushed back.

According to Cloudflare's infrastructure data, the share of 403 "Forbidden" responses served to agents and bots has roughly doubled since mid-2024. Major retailers — Amazon chief among them — are not waiting for the industry to self-regulate. They are actively blocking agent traffic at the gate. The web wasn't built for agents, and the web is actively reminding agents of that fact.

This isn't just a policy story. For every team building AI agent systems that touch commerce, payments, or user-facing purchasing flows, this shift has direct architectural consequences.

Why Retailers Are Blocking Agents — And It's Not Just Protectionism

The instinct is to frame retailer pushback as incumbents protecting turf. That's real, but it's incomplete. Three distinct pressures are converging simultaneously.

Disintermediation of ad revenue. Large ecommerce platforms derive substantial income from on-site advertising. An agent that routes a buyer directly to a product skips every sponsored placement, every upsell surface, and every behavioral data point that feeds the ad auction. From a retailer's perspective, an agent-shopper is a revenue-negative visitor.

Fraud and dispute abuse are already spiraling. Return rates and total return costs have roughly doubled and tripled since 2019, according to National Retail Federation data. The most prevalent fraud vector is dispute abuse — a customer purchases, then disputes the charge with their card issuer, leaving the merchant holding the loss. Agents introduce a plausible new excuse layer: "my agent bought that, not me." Merchants who are already bleeding on chargebacks have every reason to preemptively exclude autonomous buyers.

Liability is genuinely ambiguous. When an agent misinterprets a user instruction and purchases 10,000 wholesale units of the wrong item, who owns the return? The consumer? The agent developer? The platform that allowed the transaction? That question has no settled answer in 2025, which is exactly why teams building production commerce agents need to read our breakdown of how agent liability frameworks are evolving before shipping anything that touches a checkout.

Working on an agentic commerce integration and hitting infrastructure walls? Talk to our team about your project.

The Consumer Trust Gap Is a Bigger Problem Than the Blocks

Here is the more inconvenient finding for agentic commerce advocates: even if retailers opened their doors tomorrow, consumers are not ready to walk agents through them.

Survey data shows approximately 25% of consumers would allow an AI to purchase a range of everyday goods on their behalf. But 42% say they would not trust AI to make purchases for them at all. That is a larger trust deficit than most agentic commerce pitches account for.

Clickstream data from Similarweb adds texture to this. The majority of AI-assisted shopping sessions occur in what researchers call the "middle phase" — after discovery, before purchase. Consumers are using AI as a research accelerator: they want help narrowing options and validating price, but they want to pull the trigger themselves. AI is a consultant, not a buyer.

This has profound implications for where agentic value actually lives right now. The opportunity is not in replacing the purchase decision. It is in compressing the research phase, surfacing better comparisons faster, and handing off to a human at exactly the right moment. Teams that architect their AI development services around that handoff — rather than around full end-to-end automation — are building for the market that exists, not the one that's hypothetical.

What the TikTok Shop Data Reveals About Human-Optimized Commerce

One data point that cuts against the "agents will replace shopping" thesis is the extreme skew in TikTok Shop performance. According to Marketplace Pulse, roughly 60% of TikTok Shop sales are driven by the top 1% of sellers, and the top 10% capture approximately 93% of total sales volume. The dominant lever is video output — sellers generating 1,000+ videos command a disproportionate share of revenue.

This is a purchasing environment that is almost entirely human-optimized. It runs on parasocial trust, creator personality, and the live or near-live feel of video commerce. An agent navigating TikTok Shop to find the best price on a blender misses the entire value structure of the platform.

The implication is not that agents can never penetrate social commerce. It is that the current architecture of high-performing ecommerce — particularly in the social layer — is built around human attention in ways that agent-first design would have to fundamentally reconstruct, not merely plug into.

Building Agents That Work in a Blocked Web

For practitioners, the 403 rate doubling is a signal to build with more resilience, not to retreat from agentic architectures. A few principles we apply at NerdHeadz when shipping commerce-adjacent agents in this environment:

Respect robots.txt and rate limits as first-class constraints, not edge cases. Agents that treat access permissions as obstacles generate legal exposure and accelerate the blocking arms race for everyone.

Design explicit human-in-the-loop checkpoints at the purchase confirmation step. This is not a concession to the trust gap — it is good product design given where consumer confidence actually sits. Agents that ask for confirmation before spending money will outperform agents that spend autonomously, measured by user retention and dispute rates.

Build attribution infrastructure now. If a consumer discovers a product via an AI session, researches it in a second session, and purchases in a third, the attribution chain is broken by default. Commerce teams that solve this problem — linking AI-assisted research sessions to downstream conversions — will have a structural data advantage as the agentic layer matures.

Ready to build? NerdHeadz ships production AI in weeks, not months. Get a free estimate.

Agentic commerce is real, but the path to production is narrower than the hype suggests. Retailers are blocking at record rates, consumers still want to own the final purchase decision, and the most successful current ecommerce formats are deeply human-optimized. The builders who win in this environment are the ones designing agents for the research-and-handoff layer — not the checkout — while laying the attribution infrastructure to capture value when consumer trust eventually shifts.