AI Reading Companions Done Right: Let the Model Read With You, Not for You

After 2025, “AI reads for you” became a popular pitch. Upload a PDF, get a summary in 30 seconds — key points extracted, highlights delivered. Sounds incredibly efficient. But think about what’s missing from this process.

What’s missing is you. When you receive an AI-generated summary, you don’t know which original passages it was derived from. You don’t know what the AI left out. You don’t know how you would have reacted to those passages yourself. You’ve received a second-hand conclusion, but you’ve bypassed the process of forming your own judgment. And the value of reading lies precisely in that process — the confusion, the connections, the questioning, the revision.

Yomitomo’s stance on AI is unambiguous: AI is a reading companion, not a reading replacement. AI joins the reading you’re already doing — it doesn’t produce output in your absence.

Text Anchors: AI Responses Must Land on Specific Passages

Yomitomo’s AI features have a hard constraint: every AI-generated thought or response is bound to a specific text highlight.

This means AI can’t float up and say “this article makes three key points.” It can only respond after you’ve selected a passage, written down your reaction, and invited it into your discussion. Its response references the original text, your existing notes, and the surrounding discussion context.

The result: you cannot skip reading and jump straight to AI output. You must highlight first, write something first. And when you review later, every AI response has a clear “source location” — it’s not an abstract claim floating in the air, but a traceable judgment anchored to the original text.

Role-Based Assistants: Not “One Generic AI,” but “Multiple Specialized Reading Perspectives”

Yomitomo doesn’t give you a single “universal assistant.” It provides a set of AI roles, each with a distinct expertise. When you type @AssistantName in a discussion area, you’re inviting a specific perspective into the current conversation:

Reading Assistants lean toward companionship, explanation, and structural guidance:

@June Hartley — Margin Reading Companion: clarifies concepts and provides background context
@Gideon Frost — First-Principles Reviewer: examines premises and causal chains
@Maya Brooks — Question Mentor: transforms vague confusion into precise questions
@Marcus Reed — Insight Editor: extracts transferable insights
@Iris Chen — Concept Translator: explains terminology and conceptual lineage
@Daniel Park — Structure Navigator: identifies article structure and paragraph functions

Review Assistants lean toward evidence, logic, and clarity:

@Arthur Whitfield — Evidence Archivist: verifies facts and data provenance
@Simone Carter — Logic Auditor: detects reasoning flaws
@Victor Tan — Risk Examiner: flags overgeneralizations and hidden risks
@Julian Cross — Final Copy Editor: compresses redundancy and sharpens expression

You won’t summon all assistants at once. Which one you choose depends on the specific problem you hit with that highlight — don’t understand a concept? Call Iris Chen. Doubt a causal claim? Call Gideon Frost. Feel like you’ve written too much fluff? Call Julian Cross.

Always On-Demand, Never Intrusive

Yomitomo’s AI features are entirely optional. You can disable all assistants and use only the annotation and discussion features. You can enable just the two or three you need most. AI never jumps in unprompted — you must explicitly invite it with @AssistantName.

Underlying this design is a conviction about reading: the most valuable output of reading is not the conclusions AI produces for you, but the judgments you form when you’re stimulated, questioned, and challenged during the reading process. AI is just a tool that makes it easier to arrive at those judgments.

Who This Is For

Deep readers who want to use AI to improve reading quality without surrendering their agency. If AI summaries make you uneasy, Yomitomo’s design philosophy may be closer to how you want to read.