We have discussed it as an abstract idea regarding AI and agents. Yet, researchers are now moving closer to making it a reality.
There will come a time when shopping is handled autonomously for us. We will see agents taking over. Ultimately, it might be done in conjunction with a series of connected devices such as our refrigerators. However, we are seeing the basics of AI agents that can shop on our behalf being laid out.
Large Language Models (LLMs) are improving every few months. While this is not necessarily the pathway to AGI (or ASI), they do hold great value. We are starting to see the potential emerge as advanced research is revealed.
In other words, the chatbots we are dealing with today might only be the bare minimum of what is possible.
How likely are you to buy something?
This is a question that marketers and vendors wrestled with for centuries. Commerce is the merging of a buyer and seller. The latter is always interested in how to attract more of the former.
Over the years, marketing was the pathway. Tied into this is research designed to hone in on buyer's preferences. In the digital age, finely tuned algorithms consumed the data we provided through out online actions to provide more targeted ads. The goal was to increase the rate of commercial success.
At the core of this relationship was human buyers. While the selling became automated in many instances, that part of the equation did not change.
For example, we can see how the algorithm feeds us suggestions on Amazon. When making a purchase, it suggests complimentary products. If a person chooses to add that item, it was automated selling. Nothing was done on the part of the seller (other than an automated product placement).
What happens when the buy side gets automated?
This is what researchers are working on.
We are going to see a time when these AI systems know us better than we know ourselves. Research is taking place to design models that capture our likes and dislikes.
In other words, there will be a time when agents will decipher our intentions and handle the purchases for us.
Researchers at the University of Mannheim and ETH Zürich have found that large language models can replicate human purchase intent—the “How likely are you to buy this?” metric beloved by marketers—by transforming free-form text into structured survey data.
This will make targeting even more precise. It will also help to facilitate more efficiency in marketing.
The finding could reshape how companies conduct product testing and market research. Consumer surveys are notoriously expensive, slow, and vulnerable to bias. Synthetic respondents—if they behave like real ones—could let companies screen thousands of products or messages for a fraction of the cost.
All of this, however, could be a stepping stone to something larger. While LLMs focus on language (its in the name), the belief is that we can elevate this to understanding of attitudes. It is based on the thesis that enough data provided will lead to a replication of human judgment.
It also validates a deeper claim: that the geometry of an LLM’s semantic space encodes not just language understanding but attitudinal reasoning. By comparing answers in embedding space rather than treating them as literal text, the study demonstrates that model semantics can stand in for human judgment with surprising fidelity.
Many will dispute this. It is a repeat of the pushback I got a couple years ago when I said these models will "create" and people stated that could only be done by humans. Now we see these models use to generate songs or stories for video.
That said, we are still a long way from this outcome. What I predict is this breakthrough will unleash a flurry of research in this area. The premise of a "like/dislike" system is only one step. It is likely that other expand upon this, accelerating development in this area.
The point is that two years can pass quickly. That is a lifetime in the AI world. Bottlenecks keep being mentioned as likely to hinder progress yet, with each one, we rapidly surpass it.
For now, it is not unreasonable to think the pace will continue for at least another couple of years. This means we are looking at a much different AI world come early 2028.
Will AI shopping agents be available?
