Hark needs negotiated merchant access
Hark
The real bottleneck is not whether an agent can click through a checkout flow, it is whether major merchants let that agent stay connected. Hark is trying to turn the open web into a usable transaction layer for an AI assistant, but retailers can already detect agent traffic, challenge it, or cut it off, which pushes the model toward partner integrations that look more like Instacart, travel booking, or wallet acceptance than pure screen control.
-
Amazon made this logic explicit in November 2025 when it said third party purchasing agents should operate openly and respect merchant participation decisions, and disclosed repeated requests for Perplexity to remove Amazon from Comet. That frames agent commerce as a permissioned channel, not a free for all browser surface.
-
Website infrastructure is moving the same way. Cloudflare split AI traffic into search, training, and agent categories in July 2026, and said new defaults for some sites would block agent activity on ad supported pages starting September 15, 2026. Merchant sites now have cleaner tooling to identify and suppress browser agents at scale.
-
That does not kill Hark’s product, it changes the winning motion. Its AT&T retail and billing partnership already shows that distribution can come through negotiated access points. In commerce, the equivalent would be merchant APIs, affiliate rails, preferred checkout integrations, or whitelisted agent identities that preserve conversion and customer service quality.
The market is heading toward licensed agent commerce, where the most durable products combine automation with merchant consent, identity, and economic alignment. If Hark can package its assistant as a traffic and conversion source that merchants want to approve, its web automation layer becomes more defensible and much easier to scale across large retailers.