When Apps Start 'Talking': Two Directions for Tool Products Seen in Readwise 2.0

The most valuable information in this article isn't the fact that Readwise 2.0 added the "chat with highlights" feature itself, but rather the debate between two product routes it reflects.

Several app updates mentioned in this issue of SSPai's review seem independent when viewed separately, but put together, they reveal a clear trend. Readwise 2.0 allows you to ask questions about your highlighted notes, letting AI find answers from your book excerpts; Mactracker 5.0 added device spec comparison, letting you compare two Macs like picking phones in an Apple Store; Snapseed upgraded de-haze and color tools, using algorithm analysis to automatically adjust local areas. These three things are essentially pointing in the same direction: Tools are shifting from "static libraries" to "dynamic conversationalists."

First, look at Readwise 2.0. I ran a demo of this feature a week ago. The scenario was: You previously read a book on organizational change, made 50 highlights, and now want to quickly review "the role of middle managers in change." You don't need to flip through directories, notes, or tags; just ask a question, and AI pulls out the relevant three highlights with a contextual summary. This experience is closer to the promise of "conversing with your past self" than any "Second Brain" product.

But what I'm saying isn't whether Readwise 2.0 is good to use, but that the route it chose is fundamentally different from the route Snapseed chose—they represent two completely different product philosophies.

Viewed from three dimensions:

Interaction Mode. Readwise 2.0 shifts interaction from "browsing" to "conversation." You no longer need to know which folder or tag your notes are in; you just need to know what you want to ask. This essentially changes "user-initiated retrieval" to "AI-initiated response." Whereas Snapseed's de-haze and color tools, although also using AI, maintain an interaction mode of "user selects tool - tool executes - user adjusts parameters." One is finding answers; the other is adjusting parameters.

User Intent. The implicit assumption of conversational interaction is: Users know what they want but don't know where it is. The implicit assumption of parametric interaction is: Users know what the tool can do but don't know how to do it best. These two paths correspond to completely different user scenarios. Readwise faces the anxiety of "I've read a lot but can't remember," while Snapseed faces the compromise of "My photo isn't great, but I want to fix it."

Product Moat. Readwise 2.0's moat is data accumulation. The longer you use it, the more notes you have, the richer the context AI can access, and the more accurate the answers. Once this flywheel starts spinning, it's hard for new users to catch up. Snapseed's moat is the algorithm itself—capabilities like de-hazing and color analysis are indeed strong with Google's algorithms, but competitors can imitate them anytime. Tool-type products naturally have shallower moats than data-type products.

The core contradiction lies here: Conversational interaction makes tools "smarter," but users must pay trust—Are you willing to hand over your note data to an AI to "understand"? Parametric interaction gives users more "control," but at the cost of learning and memory burden.

I propose a framework to understand this trend: The shift of "cognitive load" in tool products.

Traditional tools (like Photoshop, Excel) place cognitive load primarily on the user—you have to learn layers, formulas, filters. AI tools (like Readwise 2.0, ChatGPT) shift cognitive load to the machine—you don't need to understand NLP or vector databases; you just need to know how to ask questions. But the cost is losing control over the process. You don't know why AI picked these 3 out of 50 book excerpts; did you miss more important information?

This trade-off is particularly evident in Mactracker 5.0's comparison feature. It didn't use AI to "recommend" which computer to buy, but listed the differences between two products, highlighting inconsistent parameters, letting you judge yourself. This is a "compromise route"—the machine helps organize information, but decision-making power stays with you. This approach is more mature than Readwise 2.0's fully automated conversation and smarter than Snapseed's pure parameter adjustment.

Returning to the initial question. Readwise 2.0's "conversational notes" and Snapseed's "auto-analysis retouching" appear to be functional differences on the surface, but are actually two routes: one is "AI thinks for you," the other is "AI reduces your operational burden." The former changes "how you acquire information," the latter changes "how you use tools."

For product managers, this might be the most fundamental fork in the road for the next five years. Are you building an assistant that thinks for the user, or a tool that enables users to think more efficiently?