Written by Brook Schaaf

The Organized Mind, a book I recently listened to (yes, hope springs eternal), might have better been called The Externalized Mind. Its main suggestion is to externalize memory using analog and digital tools like index cards, hanging file folders, lists, calendars, scanning, naming conventions, and so forth. In other words, you offload to deal with information overload.

We might say that the successor tactic shall be to offload actions. As most people don’t have an assistant, agentic AI might be just the thing. Among other examples, 

  • Google announced an Agent2Agent (A2A) protocol 
  • Amazon announced a new “Buy for Me” feature 
  • firmly.ai announced an integration with Perplexity to “scale access to Perplexity’s native shopping experience” (something Perplexity’s search engine is apparently unaware of).

Marie Haynes Consulting argues that this technology will “radically change the web.” Indeed, a five-minute demo of a Google integration is quite impressive and worth watching. It includes image recognition, associated product recommendations, and price negotiation.

With “Buy for Me,” a shopper can request that Amazon makes the purchase on their behalf. The release includes an example query for “X brand of women’s leggings,” though my own searches for Louis Vuitton, Rolex, Nike, Patagonia, Canada Goose, Dyson, and Sonos on the web and mobile apps didn’t make the button appear.

Neither did I have any luck finding the “Buy” button on Perplexity after multiple queries, despite it telling me “You can buy a wide variety of products through Perplexity’s AI-powered shopping feature, especially if you are a Pro subscriber in the US.”

I did, however, catch a glimpse into some of the problems agentic AI is likely to encounter and why I’m skeptical it will turn out as anticipated. 

  • Query Friction – incomplete information, limited merchant participation, slowness (one Perplexity query timed out), security risks 
  • Confusion and Ambiguity – intent, product purpose (document vs. picture printer paper), size, color, type, quality, brand, acceptable substitutes 
  • Fake Signals – fake reviews, fake prices (think surcharges, sizing games, unbundled components, rebates), shipping speed, associated services like installs

Thus, contrary to the sentiment that agentic AI will make buying more passive, the Google demo shows the buyer will need to be much more — not less — engaged with the seller, perhaps even with multiple sellers. In the demo, the shopper engages with the bot on the merchant site. The bot then escalates to a human supervisor to resolve a negotiation.

It’s also easy to imagine a “buy” bot squabbling with a “don’t buy” bot, each making arguments to persuade the shopper, who must now weigh more decisions and more information, some of which is likely to be wrong.

If agentic AI improves the user experience and the merchant yield, it’s likely to be embraced by both. For their part, affiliates may succeed with their own agents better than large search engines because they’ll have better-structured data and authority specific to their domains.

However, if it shifts the burden from clicking and comparing to clarifying and correcting, it may be onloading more than it is offloading. In this case, the best agent is probably still you.

The post AI Agents Lost in Space appeared first on FMTC.