Sitemap

Agentic Shopping

2 min readAug 8, 2025

--

Press enter or click to view image in full size
Image generated by AI using OpenAI’s DALL·E, prompt by the author.

Last week, I tried out ChatGPT’s Agentic mode and it was odd but inspiring. Just like trying out full self drive capability, the Agentic mode had me watching over its shoulder, so to speak, wondering what it would do next.

The task: add a bunch of items to my shopping cart at my online grocery store. It pulled up the website of the store, had me put in my credentials and go through the MFA sequence, then it proceeded to move the curser around and figure things out. What was nice was that it displayed the logic of what it was doing and also any mistakes it perceived itself making, like adding too many items to a cart.

I mostly enjoy grocery shopping. This might come down to my hunter-gatherer DNA. However, I dislike repetitive tasks and adding the same staples I get every week. I like looking for the new food types or the deals.

After a few minutes on its own, the AI agent managed to navigate the site and place most of what I wanted in the cart. I then spent a fraction of the time I’d normally spend on the site making slight tweaks before ordering. With more confidence in the system, I could have left it on its own to save even more time.

The dilemma is how much of an upfront investment do I need to make in order to get the agent to do something helpful — where I extract more value than the effort I put in. I have a concierge service available on my credit card, but I haven’t used it in over a decade because of the time it took to explain what I wanted and the mediocre results I derived from the concierge’s work.

However, ChatGPT’s Agentic Mode is less of a black box than the concierge service. I can see it get to work immediately and can correct it when it goes astray. If the assignment is properly structured, I can derive benefit even if the task isn’t fully completed. It also isn’t socially awkward. I can correct it without feeling like I’m taking advantage of a human.

Like many others, I presume that I haven’t fully grasped yet how to employ this new Agentic mode. It is hinting at being an incredibly powerful tool that can help us get access to more information to make better decisions, especially when gathering this information involves many repetitive tasks.

--

--

Leor Grebler
Leor Grebler

Written by Leor Grebler

Independent daily thoughts on all things future, voice technologies and AI. More at http://linkedin.com/in/grebler