AI shopping agents could become the new gatekeepers of online retail
WirtschaftsWoche argues that AI agents which search, compare and buy for us may decide which sellers get seen, raising a new kind of competition problem.
The short version
- Agentic commerce means AI agents search for products, compare prices and sometimes decide what to buy.
- Upside: shoppers save time, find cheaper offers and benefit from sharper price competition.
- Downside: agents could favor some sellers, sideline others or control access to the market.
- Competition could shift from winning attention to winning the agent's choice and the user's trust.
- The foundation-model market leans toward concentration, but cheaper training and specialization push back.
Why it matters
If a few agents end up doing most of our shopping, they decide which businesses we ever see. That affects prices, the choices you get and whether small sellers can reach customers at all. Regulators will have to protect the convenience without handing out another round of gatekeeper power.
The argument in brief
As WirtschaftsWoche reported in an analysis piece, the spread of artificial intelligence is creating new problems for competition policy, and the problems differ by market. The most striking one concerns AI agents, programs that act for a user rather than just answering questions.
The use case is agentic commerce. An agent looks for products on your behalf, compares prices and may even make the purchase decision. This is an opinion-driven analysis, not news of a specific launch or deal, so treat it as a framework for thinking rather than a report of events.
Why shoppers might win
The article does not dismiss the benefits. An agent that does the legwork saves people time. It can find cheaper offers more easily than a human scrolling through tabs.
Intense, automatic price comparison could also sharpen competition between sellers. In that picture, the consumer comes out ahead.
How a helper becomes a gatekeeper
The risk appears when many people rely on the same agents. Such agents could turn into powerful intermediaries standing between shoppers and sellers.
Today a business mainly has to persuade customers. In an agent-driven market, it would first have to be picked up by the algorithms. An agent could favor certain sellers, put others at a disadvantage or control who gets into the market at all. The author calls this the threat of a new generation of digital gatekeepers, meaning companies that control the entrance to a market.
The competition for attention that defines e-commerce today could turn into a competition for the agent's decisions and for the user's trust in their agent. Whoever wins that trust shapes what millions of people buy.
The market underneath the agents
The piece also looks at the market for foundation models, the large general-purpose AI systems that agents run on. It names three forces that favor concentration.
- High upfront costs. Training a strong model needs huge investment in data centers, chips, data and skilled staff, and much of it cannot be recovered. Once a model exists, serving one more user costs relatively little, so size pays off.
- A feedback loop. More users produce more usage data, which improves the product and attracts more users. Training data is not necessarily exclusive, but the combination can be a real edge.
- Vertical integration. Microsoft, Google and Amazon own models and also cloud infrastructure, sales channels, operating systems, office software and customer access. That lets them build their models into existing ecosystems very efficiently.
Reasons monopoly is not a given
The author names two counterweights. First, better training methods, smaller specialized models and improved algorithms could make strong models much cheaper to build. DeepSeek is cited as proof that very capable models can be developed without the vast resources of the biggest US tech groups. The technology also moves fast enough that today's advantages may not last.
Second, different models can be better at different jobs, such as programming, medical applications, legal research, scientific work or customer communication. A recent study by industrial economist Mert Demirer and co-authors found that the supply of commercial models is growing strongly, prices for inference are falling sharply and no single model dominates every use. Inference is when an already-trained AI uses what it knows to answer requests.
The quieter danger: lock-in
Even without a monopoly, the article warns of serious competition problems. When a company shapes its processes, data and applications around one AI model, it creates switching costs. The deeper the integration, the harder it is to move to a rival.
Customers who depend on their current provider can, over time, be exploited by it. That is why the author calls the portability of training knowledge between models hugely important for competition policy. By contrast, the market for computing capacity and AI infrastructure has high fixed costs but likely weak network effects and low switching costs. For now, the author expects effective competition there.
What comes next
The challenge, in the article's words, is to keep the efficiency gains of AI agents without allowing even more powerful gatekeepers. The piece does not propose specific rules, so what policy would look like is an open question.
For businesses, the practical reading is that visibility to agents may matter as much as visibility to people. For shoppers, it is worth asking whose interests an agent serves when it picks the winner.
What to watch
- Whether regulators propose rules on how shopping agents rank and choose sellers.
- Whether switching between AI models becomes easier or harder for businesses.
- Whether cheaper, specialized models keep the foundation-model market from concentrating.
Try it
Next time an AI assistant compares products for you, check how many sellers it shows and ask it to explain why it ranked the top pick first.
Sources and links
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by WirtschaftsWoche · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
