Reading, together

A slow-burning idea

I have a long list of things I want to read, and the order I read them in is basically arbitrary. What I've never had is a low-friction way to read them with someone.

Two walls I keep hitting

Every time I've actually tried this — at P2PU, in graduate seminars, in ad hoc groups — I run into the same two walls. And the cruel thing is that they push against each other.

The first wall is commitment. The moment you ask people to sign up for "ten weeks, every Tuesday," half of them are gone by week three, and once a few drop out the rest feel slightly silly still showing up. I've watched this happen over and over. And the honest reason people stay, when they do, has almost nothing to do with the reading:

The second wall is preparation. If people haven't done the reading, the conversation is hollow. This is the part the tools-and-platforms people always underrate: any socially-based pedagogy — a seminar, a book club, knowledge-building — is only ever as good as the preparation sitting underneath it. Four clever people who skimmed the thing on the train will have a perfectly pleasant chat and learn very little. (I have a note to myself about book clubs that I keep coming back to: the danger is that you spend the whole evening talking about everything except the book — the book becomes a thin alibi.)

And here's the trap: the cure for the second wall — serious, shared, genuine preparation — makes the first wall worse, because it asks for more commitment, which produces more drop-out. Push on depth and you lose people. Lower the commitment to keep people and you lose the depth. Almost everything I've tried has fallen into one ditch or the other.

Grab a coffee with a stranger

The thing that cracked this open for me was completely mundane: those "grab a coffee" networks that pair you with a random person for thirty minutes.

I did a bunch of these — the ones where you opt in and get assigned a thirty-minute call with some random founder or researcher — and had genuinely interesting conversations. But what really struck me was how little it took to run. It's just scheduling. No cohort, no curriculum, no real platform — you opt in, you get matched, you talk, you're done, and there's zero ongoing commitment. Nobody drops out of a single coffee. The friction is almost nothing.

And I kept thinking: I have a hundred academic articles I'd love to discuss with one other person — why isn't there a version of this for reading? Click a paper, get matched with whoever else wants to read it this week, half an hour on the calendar, done.

The obvious problem, of course, is that pure random pairing is shallow. Two strangers who happen to both be free on Friday can have a nice chat, but they can't go deep, because they've built nothing together — no shared vocabulary, no shared ground. The very thing that makes the coffee network frictionless (no history, no commitment) is what caps how far it can go.

The thing the seminar gets right

At the far opposite end is the thing I actually love — and can almost never fit into my life.

A good graduate seminar, or something like Zena Hitz's Catherine Project (free, online, close reading of serious books, led by St. John's-style tutors), gets the one thing the coffee network can't: by the time you're a few weeks in, everyone has read the same papers, and you can finally have the conversation that only becomes possible on top of shared ground. You're not re-explaining the basics; you're building. It's wonderful, and it's the kind of depth I keep chasing.

But it's also exactly the high-commitment, same-people-every-week model — which is to say it sits squarely behind the first wall. It works beautifully for the small number of people who can clear ten Tuesday evenings in a row, and it's out of reach for almost everyone else.

So those are the two poles I'd been bouncing between for years. Low-friction-but-shallow, or deep-but-rigid. The question that wouldn't leave me alone: could you get the easy scheduling of the coffee network and the shared context of the seminar — without forcing the same people to stay together for ten weeks?

Project Fluid

This is the idea I keep coming back to, and the one I've never actually built. Everything else on this page is the soil it grew out of; this is the seed I still want to plant.

The trick, I think, is a dependency graph — a little knowledge map of readings, where each one sits on top of the ones before it. You unlock a level by reading the paper and discussing it with someone — but, crucially, it doesn't have to be the same someone each time.

Say I'm reading a paper at level four. I get matched with a complete stranger, or with you, or with a couple of people — and the one thing I know for certain is that whoever I'm paired with has already worked through levels one, two and three. (A "level" needn't be a single paper, either — level one might be three of them.) The shared context is guaranteed by the structure of the graph, not by everyone heroically showing up every week. The pairing stays as easy as booking a coffee; the depth comes for free from the prerequisites.

That's the whole move, and it's what dissolves the tension between the two poles. You're never committing to a sequence with a fixed group, so the drop-out wall mostly disappears — you move through the graph at your own pace, with whoever's around when you're ready. And yet every conversation happens between people who share real context, so it can actually go deep.

Inside each step, a small loop:

1. Read the paper.
2. Write a short response before you meet — partly to prepare your own thinking, partly as an honest check that you actually did the reading.
   ↑ you can't book the conversation until it's written
3. Meet for thirty minutes — one person, or a few.
4. Compile something tiny at the end. One insight. Three good questions. It can be almost nothing.

The written response is doing quiet double duty: it's how you arrive with something to say, and it's the gate that handles the second wall — if the price of admission to the conversation is having written your own short take, you can't show up having skimmed it on the train. (And there's something I genuinely believe about the order here: it's valuable to wrestle with a paper from first principles before you see what everyone else thought. Read the Piaget paper and figure out what you make of it, rather than reading fifty critiques of Piaget first. The gate protects that solitary struggle, and only then opens the door to other people's connections.)

What everyone leaves behind

Here's the part I find most interesting, and the part I'm least sure how to get right. Those tiny end-of-meeting compilations add up.

If a lot of people move through the same pipeline, then after a while every paper in the graph has accumulated a real body of insights and questions — one pair's three good questions, another's single hard-won insight, and another's, and another's. The question that's kept me up at night is: how does the work of all those previous readers benefit you — without robbing you of the chance to think it through yourself first?

This is crowdsourcing, but I don't think crowdsourcing has to be superficial if you design it right. The move I keep reaching for is sequencing: you only unlock everyone else's notes after you've done your own reading, written your own response, and had your own conversation. So you get the first-principles struggle, and then the enrichment — oh, someone connected this to barefoot schools in India, someone else to a fireman's academy — in the right order, instead of having your first encounter with the paper pre-chewed.

It also quietly fixes something that's always bothered me about wikis. Writing on Wikipedia used to be a way of learning; now, for most topics, you feel you've got nothing to add unless you're already an expert — you can learn by reading, but you can't really learn by creating anymore. A fresh paper with a growing pool of prior readers' questions is the opposite: there's always room for your contribution, because the pool is built precisely from people at your stage working it out. (There's plenty more to figure out here — you could imagine A/B-testing which discussion prompt produced the best conversation, or layering spaced-repetition cards on top, à la Quantum Country — but that's all downstream of getting the basic accretion right.)

The interdependency problem

There's an older version of this puzzle in my PhD thesis, and I think it's actually the heart of the whole thing.

My thesis studied a massive open online course — a few thousand people — where a small number contributed intensely, really went deep, while a much larger group hovered around the edges, reading along but rarely posting. (In the literature this gets called "legitimate peripheral participation," and one of my favourite moments was discovering, months after a course I'd run, a Brazilian participant who'd followed the whole thing closely and learned a huge amount without any of us organisers ever knowing he existed.)

The design question that fascinated me then is exactly the one Project Fluid runs into now: how can the people who go deep benefit from having a big surface community around them, and how can the surface people benefit from the few who go deep? Get that interdependency right and the whole thing becomes more than the sum of its parts. Get it wrong and the deep contributors feel like they're carrying dead weight, and they leave.

What I like about the Fluid mechanism is that it tries to build the mutual benefit into the structure rather than hoping for it: the people who go deep leave artefacts that the next person literally stands on, and the next person's short response feeds straight back into the pool that the deep readers will later draw from. The interdependency stops being a matter of goodwill and becomes a matter of how the graph is wired.

The soil

Everything else I've chewed on over the years is really just soil for that one idea. A few of the threads, briefly, in case any of them is useful on its own.

Divergence and convergence

I gave a talk in 2011 — Grappling with ideas — about how we actually work with ideas, alone and in groups. The core of it: good thinking runs in a cycle of divergence (generate freely, throw everything onto the wall) and convergence (step back, cluster, name what emerged, then dig deeper into a named cluster). A reading group lives or dies on this. The reading and the written response are divergence; the conversation is convergence. The dead version — what I called stimulus/response — is the seminar where everyone posts a comment to be seen posting and nothing ever gets gathered up. (Which is why, watching grassroots Vervaeke groups, I noticed that people just summarising the content learned very little; what helped was when they went past the content to ask questions, make connections, try to apply it.)

Artefact and discourse

A distinction I've leaned on for years: the difference between the conversation (discourse — cumulative, bound to its moment, like a chat or a forum thread) and the thing the conversation leaves behind (the artefact — integrative, rewritten to always represent where the group has gotten to, like a wiki page). Most platforms only capture one. The accreting pool of questions in Fluid is an attempt to let the fleeting discourse of a hundred thirty-minute calls condense into a durable artefact — which is exactly the move I once watched a Knowledge Forum class make, reorganising weeks of scattered notes into shared topical views until you could see what the group, collectively, knew.

What the notes are for

Underneath the group is always an individual, taking notes — and I've spent a long time on why that's worth doing. The surprising part is that most of the value arrives late: it's good to write something down and wait, because when you come back the time that's passed has filtered what actually mattered, re-reading reinforces the memory, and you see new connections. So a note pays off along two axes — for yourself now (the thinking-by-writing), for yourself later (retention), and for other people (the public artefact). This is the whole reason I'm building Petrarca, a reading app that does the individual half — it breaks articles into atomic claims, models what you already know, and surfaces new reading at about 70% novelty. Which, I realised with some delight, is the same Goldilocks idea I'd wanted for matching reading partners, just applied to matching articles to a person instead.

Three levels, and "learning about"

The frame I hang all of this on: working with ideas happens at three nested scales — individually (the notebook, the Zettelkasten, Petrarca), in small groups (the seminar, Knowledge Building, "what does the group know?"), and in networks (the connectivist MOOC, the edublogosphere). Project Fluid is really an attempt to bridge them — a private struggle that surfaces into a community pool. And one bias worth naming: I care about learning about, not learning to. Skills you can mostly pick up just-in-time off Stack Overflow. The slow, generative work of making sense of a hard idea with other people is the thing that's genuinely underserved — and honestly there's very little difference between that and research.

Who else has tried

I'm not the first to want this. The original MOOCs (the cMOOCs of 2008 — community-first, knowledge-creation-first, before Coursera turned the format into broadcast video) were reaching for it; P2PU spent years on it and eventually found its footing running free learning circles in public libraries; the Catherine Project is the loveliest current version; Online Great Books and Interintellect are the paid cousins. None of them, as far as I can tell, has tried the specific non-persistent-pairing-on-a-dependency-graph move — which is either a sign that it's a genuinely new idea, or a sign that everyone smarter than me already saw why it won't work.

Still just a bet

The core mechanism — the dependency graph, the changing partners with guaranteed shared context, the accreting pool of tiny artefacts — I have never actually built or tested.

Everything around it I've felt in my bones, from fifteen years of seminars and P2PU and reading groups gone right and (mostly) wrong. But the central bet is still a bet. Maybe it doesn't even need to be a platform — the Catherine Project gets remarkably far on culture and a roster of committed volunteers, and I have a nagging suspicion that the social glue matters more than any graph. Maybe the dependency graph is over-engineering, and the brute simplicity of the coffee network was the whole point all along. I genuinely don't know, and the only way to find out is to run it — even the crudest version, a Google form and some manual matching, would teach me most of what I want to know.

So I'll end the way I'd end at a dinner table, not with a conclusion. Have you ever been in a reading group that actually held? What made it stick — and what, if anything, did you do with the notes afterwards? And if you've built something in this space, or want to, I'd genuinely love to compare notes. You can find me at reganmian.net or @houshuang.