Nate Meyvis

Google's Dreambeans app

I've just gotten off the waitlist for Dreambeans, Google's new app for "proactive, personalized collections of stories each day covering the things that matter most to you." You give it access to Google services, it processes the information, and it gives you ideas for things to do. It's not, at least in my experience, any kind of general "stories" app; mine is all suggestions, mostly of things to do and partly of things to learn and research.

Each suggestion comes with a drawing, and they absolutely nailed the color palette and pencil-and-paper aesthetic of these. There are some standard AI errors: internal inconsistencies, playing cards turned the wrong way, people looking where it makes no sense for them to be looking, famous architecture depicted unfaithfully, and so on. That said, if you're interested in looking at calm, speculative, slightly whimsical pictures with elements of your life, check these out.

The pictures are the discernible evidence of AI being used effectively. The prose summaries of the suggestions are unoffensive but do not read well. The suggestions themselves are only thinly related to my actual life; there is no evidence of a suggestion engine here that trad-Google could not have achieved (and surpassed). Many of the suggestions are incompatible with my actual life in ways that were in principle accessible to the system: e.g., by misinterpreting a travel itinerary and giving correspondingly impractical suggestions.

A sort of combination of these errors is a suggestion that I generate mentorship advice for academics by "mapping cloud system architectures using formal logic:"

CleanShot 2026-06-14 at 12

This is pure AI slop in that there are various surface-level connections to my actual life, but almost nothing about it makes any sense, however appealing that cup of coffee looks. Similarly: Dreambeans is telling me to review a tutorial about foaming weed killer. But the system should know that (i) I'm done with that yearly maintenance, (ii) one needs only the slightest knowledge of the mechanics of weed killer to use it, and (iii) I already have that knowledge.

Using this app feels a bit like making small talk with a colleague who's trying to show interest in your life by keyword-matching things they once heard you say against a list of things that are on their radar. Such "suggestions" tend to be irrelevant at best or, if not, an attempt to impose their value system onto you. That feels bad. Dreambeans is presenting an aspirational picture of my life, not in having the slightest sensitivity to my aspirations but in finding ways to staple someone else's aspirational picture of life to the facts of my own.

Given my values, I cannot afford Michelin-starred tasting menus. My preferred way of life leaves me totally uninterested in "midweek escapes" (and, if I planned such an "escape" to listen to a piano recital, I would not sit with my back to the piano). I love '90s music but would never "grab the anniversary pressing" of Local H's As Good as Dead; I don't even own a record player. These "stories," and the whole Dreambeans experience, tell me much more about the value system of a Google product manager than about the possibilities of my own life.

With all that said:

  1. I'm open-minded about AI's potential to help us plan our lives. (I've read at least one wonderful novel I wouldn't have found except by asking AI for suggestions.)
  2. I'd much rather see an early version of an app like this than to wait another six months for it; who knows what future versions might do.
  3. If Dreambeans synthesizes material about my life as effectively as Gmail does, this could be really great.
  4. Did I mention the pictures look really nice?

Mechanical notes: I had attempted to take a week off, but I got the itch. I'll tack a day on to the end of my break. Here as always, I'm speaking only for myself. I am an ex-Googler and think that Googlers are an awesome crowd.

#AI for the sake of AI #Google #generative AI #sociology of software #software