The online world is set for another big shift in consumer behaviour. People who once relied on search engines, online aggregators and price-comparison sites to find shopping, travel, insurance and other deals are dabbling with AI agents to do all this for them.
Against this backdrop, Amazon’s blocking of Meta’s Muse agent is not just a dispute over bots shopping online, but a bugle blown in a battle likely to unfold over control of the internet’s agentic layer.
Amazon says Muse barged onto its site without permission, raising security and privacy flags, while Meta disputes some aspects of that charge. The latter wants Muse widely adopted as a shopper, but Amazon has its own agent being developed for that job.
As old lines of trespassing blur and Google, OpenAI, Anthropic, Microsoft, Apple and Perplexity rush to offer AI agents, a race to be everyone’s bot-in-chief could see turf wars erupt that might carve up the web.
Business models are already getting rattled as websites watch human customers replaced with hard-nosed agents. An AI agent can scan all that is on offer, analyse evaluatory data and place an order on our behalf within a flash. The website thus acts as a supplier to an agent that is in control of the human interface. Acquiring agency over bulk spending, Big AI may reckon, is worth a fortune.
As for intermediaries being cut out of the loop, it is not new; the internet did that by drawing businesses and customers closer across various sectors. Agentic commerce could now see e-com websites lose the influence they held over customer choices as online devices—like finely targeted ads—get outmoded. Unless they arm up with agents of their own.
In India, Amazon, Flipkart, Myntra, Reliance, Meesho and others could plausibly enter the agentic fray. With UPI’s operator working on a way to clear AI agents for making payments on behalf of users, India could well become a test market for a grand shift to agentic shopping. Globally, the transaction value of such e-com is pegged at about $7.7 billion in 2026, with forecasts of exponential growth.
On the flip side, regulators must guard against too many agentic decisions being made by too few kinds of bots, even if the market comes to assume a personal agent as an entity distinct from its maker. A handful of models acting as gatekeepers of what humans get to see, which merchants can reach us and potentially what we pay could spell an antitrust crisis across the internet.
Agent multiplicity, on the other hand, could multiply security risks once bots are allowed to move money around. To fortify websites, Amazon’s AgentCore Web Bot Auth is trying out a crypto ID system for AI agents; it lets a website check agents and apply its own access policy. A Fast Identity Online alliance is also aiming for safeguards. But a maze of agentic barriers may leave much of the internet walled off to agents, which would impact the value they can deliver.
Amazon’s legal battle with Perplexity illustrates how unsettled cyber- boundaries remain. In August, a US appeals court lifted an injunction that stopped Perplexity’s agentic browsing of Amazon’s site, but the underlying dispute has not been resolved.
The online retailer’s recent rejection of Muse is under watch from many angles. Agentic market capture is an important one, given how margins could move with power. As the internet’s buzz shifts from websites and apps to AI agents, we must brace for a fierce contest to own the customer’s intent by acting as a trusted doer.
