What this covers

In this conversation, Ben Springwater is questioned by Dom and Matt about building Matter, a read later app. Ben does most of the talking. The inquiry starts from why recommenders fail, through Recency and Popularity Bias and a lack of exploration, and inverts those failures into design principles. Good curation, in his framing, sits in the high-quality, nonobvious corner of a two-by-two, an application of Peter Thiel's non-consensus idea. Dom adds a wrinkle: if algorithms get good enough to surface every under-read piece, the hidden gems vanish, leaving human curators who dig in obscure corners.

The middle stretch is product craft. Ben argues that a feature's real cost includes what it does to users who never touch it, and that Matter now prioritizes getting new users to a magic moment over delighting power users. He explains why a free tier plus subscription beats ads, why a giant archive of links fails against the paradox of choice, and why parsing messy web pages remains hard. Audio gets its own thread: improving text-to-speech and transcription, readable podcasts as a listening companion, and why listeners favor a medium's native format.

The conversation then turns to writing itself. Ben discusses the Art of Appreciation as precise description, the continuum between David Foster Wallace's enthusiasm and John McPhee's restraint, and why John Gardner's wisdom writing works despite sounding banal when paraphrased. A contested detour follows: Scott Alexander's case for universal basic income over universal basic jobs, which changed Ben's mind.

Sharpest takeaway

Matter founder Ben Springwater argues that reading products should act as a curated decision engine that surfaces nonobvious, timeless writing through a hybrid of algorithms and trusted human curators, rather than the recency and popularity signals most recommendation systems rely on.

  • Recommendation systems over-index on recency, popularity, and the user's stated interests, which misses hidden gems
  • Algorithms handle concrete topic tagging well but cannot yet capture abstract relationships like tone or worldview, so human curation remains necessary
  • Subscription revenue suits a smaller, high-value audience better than an ad model that needs huge impression volumes

The argument · threads9 threads · 53 claims
0.85

Popularity, recency, and exploitation biases make recommenders miss timeless, nonobvious hidden gems.

6 pointscentrality 5/5
  • Good curation occupies the quadrant of high quality and nonobvious content on a two-by-two matrix, because high quality that is also obvious adds no value over what a reader would find on their own, which is the Peter Thiel non-consensus idea applied to reading.

    “if you're in high quality and obvious, then you're not really offering value over replacement”

  • Recommendation systems that rely on popularity by definition miss the hidden gems and needles in the haystack, which are often the most rewarding things to read.

    “that is going to miss all of the hidden gems and needles in the haystack that I think are the most rewarding things to read”

  • Most recommendation systems weight recency heavily because they key off what most people pay attention to right now, creating a strong recency bias that pushes out timeless content.

    “they index so heavily on recency. And in large part, that's just keying off of what most people pay attention to.”

  • Recommendation systems lean too far toward exploitation of known user interests, which produces boring results and corners users into their existing interests because no system has enough signal to capture the full range of what someone might like.

    “they sort of are all on the side of exploit. That produces results that get really boring, and you get cornered into interest.”

  • The three recommender failure modes can be inverted into design principles: focus on timeless rather than recent content, find nonobvious rather than popular content, and balance explore and exploit.

    “I think taking those three things, you can kind of invert them”

  • Personalization is powerful when it recommends canonical follow-ups to a topic a reader is already exploring, such as suggesting three essential essays after someone reads about microprocessors, but the system must also keep providing serendipity and range.

    “We want to continue providing serendipity and range.”

0.61

Hybrid systems pairing algorithms with vetted human curators are needed for quality curation.

4 pointscentrality 4/5
  • The most effective curation system will be a hybrid of algorithms and human curators, because the locus of trust will remain relevant and quality is best maintained by relying on people who have been vetted for good taste.

    “we think that most effective system is going to be a hybrid between algorithms and human curators”

  • Matter maintains a large pool of content above a quality bar, either recommended by trusted people or vouched for by the team, and growing this pool lets the product serve more and more niche interests, while the harder problem is distributing items from the pool to individuals.

    “we have what's essentially a big pool of content that we think is above a certain quality bar”

  • Algorithms are good at tagging content with concrete topics and mapping relationships within a topic, but they still cannot understand high-level, abstract relationships such as tone, theme, or implicit worldview, so human curators remain necessary.

    “What the computer is really good at is assigning topics that are concrete to content”

  • Matter can scale curation by finding people with good taste and feeding their recommendations into the system, creating a curate-the-curators model that produces a long tail of recommendations from trusted individuals rather than only a headline feed.

    “There's a way of scaling that up to curate the curators.”

0.57

Matter's product choices favor focused value over feature bloat, ads, and power-user catering.

6 pointscentrality 3/5
  • Evernote became bloated by building every feature requested by power users and trying to be many things to many people, which Ben cites as a classic cautionary example of product overreach.

    “Evernote is this classic example of a product that ultimately became bloated by building every single feature that was requested by every power user”

  • Matter now emphasizes getting incremental new users to a magic moment of value quickly, because moving a new user from 10 or 15 to 30 on a retention scale matters more than moving a committed user from 70 to 80, even though serving power users is the intuitive ethos.

    “we want to get them to 30 rather than taking someone who's at 70 and getting them to 75 or getting them to 80”

  • Matter chose a free tier with a premium subscription model over advertising because display ads require tens or hundreds of millions of impressions per day to be profitable, whereas subscriptions can sustain a good business with an audience an order of magnitude smaller if the product delivers enough value.

    “requires a really large audience, tens or hundreds of millions of impressions per day to build a really good business”

  • Handing users a huge archive of links fails psychologically because of the paradox of choice, so the product should deliver a regular daily amount of value, neither too much nor too little.

    “The psychology of that just doesn't work. There's the paradox of choice. It's overwhelming.”

  • The true cost of an incremental feature is not only development time and the opportunity cost of other work, but also what it does to the product for users who never use it, such as cluttering the interface, complicating it, or making basic functions harder.

    “the cost is really the like important underemphasized thing in this equation”

  • Matter aims to meet the moment of wanting something for the mind with content that is tasty, snackable, and nutritious, where nutritious means time well spent, and where tasty and snackable are hard to find in much really good content.

    “we think that the way to meet that moment is with content that is tasty, snackable and nutritious”

0.53

Audio is a native, uniquely suited medium for voice, enthusiasm, and practitioner expertise.

5 pointscentrality 3/5
  • Most podcast listeners prefer audio because podcasts are native to audio, and the voice, tone, and interplay of guests is part of the content itself, not just the words, so there will always be a bias toward the original modality.

    “Podcasts are a medium that are native to audio. There's something about the voice and the tone and the interplay of the guests that is really part of the content.”

  • Writing at high depth takes a long time, so it is harder for practitioners with full-time jobs to write, whereas audio podcasts let practitioners share their knowledge in an hour or two, making audio a uniquely suited format for disseminating practitioner expertise.

    “practitioners can impart a lot of their knowledge in just an hour or two”

  • Improvements in text-to-speech and transcription, which are inverse technologies, are getting close to human-like accuracy, so Matter can let users seamlessly switch between listening to written content and reading audio content.

    “the technologies that were powering text-to-speech and transcription, the inverse, were getting better. They were getting more accurate and more human-like.”

  • Matter's readable podcasts feature is meant to be a companion to listening, producing a time-synced transcript so users can capture ideas during playback, rather than a way for people to replace listening with reading.

    “our intention with readable podcasts was to create a companion to that primary listening experience”

  • Podcasting is a strong medium for expressing appreciation and enthusiasm because those qualities come through in the voice, and conversational formats build on the energy of guests and present the best version of them, which is why shows like Acquired and Founders appeal to listeners.

    “I think it does a really good job expressing appreciation and enthusiasm and being fans of things and admiring things just because that can come through in the voice.”

0.47

Matter grew from weak reading tools and the value of manual curation labor.

4 pointscentrality 3/5
  • Email inboxes are a poor place to read newsletters, and no product had integrated newsletters into a dedicated reading environment, which created an opening for Matter.

    “Your inbox isn't a great place to read, and no product integrated your newsletters into the place that you went to read.”

  • Ben Springwater argues that existing reading apps are long in the tooth, and that the importance of reading in people's lives makes the gap between weak tools and that importance puzzling, which motivated Matter's founders to build a better reading tool.

    “we felt like the state of reading apps just wasn't very good. These tools have gotten kind of long in the tooth.”

  • Ben Springwater's idea for Matter came from his hobby of sending a monthly roundup of the best things he read to friends, which showed him that curation has real value and that the labor of finding good content is worth offloading from others, and the founders asked whether this manual process could be scaled, productized, and personalized into a better decision engine for daily reading.

    “there is value in curation, there's value in doing the labor to find good stuff, so that other people don't have to spend the time doing that search”

  • Springwater realized that the value of curation lies in doing the labor of finding good material so other people do not have to spend their time searching, and that people responded very positively to his monthly hand-built reading roundup.

    “there is value in curation, there's value in doing the labor to find good stuff, so that other people don't have to spend the time doing that search.”

0.39

Scott Alexander's basic income argument overturned Ben's view that jobs are an irreducible source of meaning.

5 pointscentrality 2/5
  • Scott Alexander argues that the basic jobs argument rests on a hidden step: that people need and enjoy work but are too unaware to realize it and will never seek the work they secretly crave unless forced, and he concludes that giving people money is preferable to a policy of universal basic jobs.

    “The argument is that people need and enjoy work, but also, they are too unaware to realize this and will never get the work that they secretly crave unless we force them into it.”

  • Scott Alexander's Slate Star Codex post on basic income, which Ben read around its 2018 publication, overturned Ben's prior mild opposition to universal basic income by dismantling his intuition that jobs are an irreducible source of meaning.

    “his post just completely demolished that intuition and flipped it on its head”

  • Alexander argues that the defining feature of capitalism is that people can get money without working, and that the desirable part of capitalism is precisely this, while the part needing fixing is that most people do not get this benefit.

    “The part where you get money without working is the fun part of capitalism.”

  • Scott Alexander has written over 1,500 blog posts while holding a full-time job as a psychiatrist, and most maintain a high level of rigor and insight, making him a singular intellectual force.

    “He also has a full-time job as a psychiatrist. But he's written over 1,500 blogposts.”

  • Finding purpose from a job is a crapshoot at best, while finding purpose outside one's job is a natural part of the human condition, and the basic jobs argument seems to worry about the opposite.

    “Getting a sense of purpose from your job is a crapshoot at best, getting a sense of purpose outside your job is a natural part of the human condition.”

0.34

Appreciation is a precise craft, and both personal and restrained writing can convey it.

4 pointscentrality 2/5
  • The art of appreciation is not just raw emotion but the skill of precisely describing the ways in which something is good or worthy of appreciation, and that precision is a craft.

    “it's being able to precisely describe the ways in which something is good or worthy of appreciation. That's a skill. That's a craft.”

  • There are too many critics and critiques and not enough fans and appreciators, and writers like Sam Anderson, who write from reverence for their subjects, help correct this imbalance.

    “by and large, there are too many critics and there's too many critiques and not enough fans and appreciators”

  • How much an author inserts themselves into writing exists on a continuum, and both highly personal, enthusiastic writing like David Foster Wallace's on Roger Federer and restrained writing like John McPhee's can be effective, because emotion can come through without first-person statements.

    “I think there is a continuum, and it's possible to do a lot of that and have it be effective and to be very restrained and have it be effective.”

  • John McPhee's restrained profile of Bill Bradley shows deep respect and affinity without first-person opinion, because the respect comes through in precise description of how the player is unique.

    “he does it in a way that's very subtle. And he's not making I statements and interjecting himself.”

0.34

Read-later users over-save and manage queues differently, and long reads fit poorly into short gaps.

4 pointscentrality 2/5
  • People save far more articles to read later than they actually read, roughly five times or more, which Ben describes as a universal law of read-later behavior.

    “it is just a universal law that people save 5x plus more than they actually read”

  • Read-later users differ in how they treat their queue: some manage it toward zero as a triage system, while others are comfortable with an ever-expanding queue that creates no mental pressure and enjoy digging through it.

    “Some people are the inbox-zero type, and they like to manage their queue down to nothing, and it's a triage system.”

  • Matter's long-form reading experience is poorly suited to short gaps in the day, such as waiting in line, because the activation energy of starting a long read does not match those contexts, and people turn to social media in those moments instead.

    “It's just the activation energy of getting into a long reading session just doesn't match the context.”

  • Building a parser that reliably produces clean readable text from the wide variety of web pages a user might save is a difficult, ongoing investment, and it is one of the core technical challenges a read-later product must solve.

    “creating a parser so that you can reliably produce very good readability is a pretty big ongoing investment for us”

0.29

Gardner's wisdom writing is persuasive through sharp language and credible, humble, high-expectation tone.

3 pointscentrality 2/5
  • John Gardner's essay The Road to Self-Renewal uses wisdom-writing to warn mid-career people that some individuals' clocks stop as they run out of steam and stop learning, urging readers never to stop learning and growing.

    “he cautions that there are people in life whose clock stops at a certain point, who run out of steam, they stop learning, they stop growing”

  • Wisdom writing like The Road to Self-Renewal sounds banal when its message is restated, because its persuasive power lies in the sharpness of language and the humble, empathetic, yet high-expectation tone of the speaker, which makes his belief in the reader credible.

    “the punch is really in the language, the power of the language, and the tone of the person delivering the language”

  • Gardner's offhand line that he is 77 and just took a new job while still learning gives the essay additional credibility because it shows he practices what he preaches.

    “But there's also that little snippet that he slips in that gave an additional piece of credibility to him that he's practicing what he preaches.”