Breakdowns: Software. Eighteen products, taken apart one decision at a time.

Warm-up: what do you already see?

Before the method, a quick diagnostic. You have used most of these products, or products like them. The point is not to score well. The point is to notice how much you already sense about why software feels the way it feels, and how often that sense is vague. By the end of the guide the vague feelings should have names.

Every piece of software is the visible end of a chain of decisions, from "what job is this for" down to "what does this button say"; reading software well means walking that chain on purpose, one lens at a time.

When you open a beautiful app, your first reaction is a feeling: calm, fast, fussy, cheap. Feelings are useful data, but they are not explanations. You cannot steal a feeling. You can steal a decision. So the first skill of a breakdown is to translate "this feels calm" into "the app hides everything except the current paragraph, uses one typeface at one size, and never shows a badge". Now you have three decisions you can test in your own product.

In Breakdowns: Objects you used seven lenses for physical things. Software needs the same seven, adapted, plus two questions that physical objects rarely force on you: how the business model shapes the screen, and what was deliberately left out. Software is infinitely extensible, so omission is the hardest and most revealing decision of all.

The lenses, briefly

Click a lens below to see the questions it asks and how it reads a product you probably know, Things 3. Each question is phrased so it has a concrete answer you could check on screen.

Mechanism over adjectives

The most common failure in teardowns is the adjective trap: "clean", "intuitive", "delightful", "modern". These words describe your reaction, not the product. Replace every adjective with a sentence that has a decision, a reason and a cost. "Clean" becomes "uses a single accent colour, so the only coloured thing on screen is the thing you can act on; cost: nothing else can draw attention, including warnings".

Why the business model gets its own lens

A free app paid for by ads must win back your attention, so it grows notifications, feeds and badges. A subscription app must justify a monthly charge, so it ships visible features and sync. A one-time purchase must win you at the moment of sale, so it invests in the first impression and in craft that reviewers notice. None of these is evil by default. But you cannot understand why Duolingo nags and iA Writer does not without asking who pays and for what. Many design "mysteries" are just the revenue model showing through the UI.

The ninth lens is the hardest: what is not there? Things has no collaboration. iA Writer has almost no formatting toolbar. Linear does not let every team invent its own workflow. Craigslist has no images on its home page. Absences are invisible unless you list the obvious features a competitor has and ask, for each one, "did they forget, or did they refuse?"

In the wild

• The Elements of User ExperienceJesse James Garrett, 2002
  Garrett's five planes (strategy, scope, structure, skeleton, surface) are the ancestor of these lenses: decisions on the lower, abstract planes constrain the upper, visible ones. Explain a surface by looking underneath it.
• Teardown writingMany designers, 2010s onward
  The most useful public teardowns (for example long-form app reviews on MacStories, or onboarding teardowns on sites that collect them) work screen by screen and name decisions. The weakest ones rate things out of ten. Read the good ones as models for your own notes.
• The Design of Everyday ThingsDon Norman, 1988, revised 2013
  Norman's conceptual model and feedback ideas are lens 2 and lens 6 in this guide. The vocabulary you already have from The Human Mind is your teardown toolkit.

A teardown is a routine, not an inspiration: capture every screen of one real task, pin each decision with a number, then write decision, reason, cost and steal for each pin.

If you just browse an app looking for good ideas, you will find the ones you already agree with. A protocol fixes that. It makes you look at the boring screens (settings, errors, empty states, the paywall) where products reveal their real priorities. It also produces an artefact you can come back to, which matters more than any single insight.

The protocol, step by step

1. Pick one task, phrased as a user would. "Pay my friend back ₹450 at a tea stall", not "test the payments flow". Write the starting state too: fresh install, or a returning user with data.
2. Capture every screen, including the boring ones. Use the system screenshot shortcut (or a screen recording, then pull frames). Number the files in order. Include loading, empty, error and permission screens. On desktop, capture at a consistent window size.
3. Make a strip. Put the screenshots in a row in any canvas tool (Figma, Keynote, even a slide). A strip shows the rhythm of a flow: how many steps, where it slows down, where it asks for something.
4. Pin decisions with numbers, not arrows. Put a small numbered circle on each decision. Arrows and handwriting clutter the screenshot; numbers keep it readable and map to a list.
5. Write the list. For each pin: decision (what), reason (why you think), cost (what it gives up), and confidence (high, medium, guess). Mark guesses as guesses.
6. Compare with one rival. Do the same task in a competitor. The differences are where the decisions become visible.
7. Steal one thing. End with one concrete change for your own product, small enough to ship this week.

In the wild

• Mobbin, Page FlowsPattern libraries, 2010s–2020s
  Libraries of real app screens and recorded flows organized by task. They are the "capture" step done for you. Useful for breadth, but you still have to write the reasons yourself.
• Heuristic evaluationJakob Nielsen and Rolf Molich, 1990
  A structured review against a short list of principles. A teardown borrows its discipline and adds "why" and "steal". See Testing & Iteration for running one on your own app.
• MacStories reviewsFederico Viticci and team, 2009–
  Long app reviews that walk feature by feature with screenshots and context on why a feature exists. A good model for writing up teardowns at length.

iA began as a design studio that wrote loudly about web typography ("Web Design is 95% Typography" is one of its well-known essays). iA Writer is that argument turned into a product. When it launched on the iPad in 2010, writing apps competed on features: styles, fonts, toolbars. iA Writer competed on absence. It opened to a blank page in a monospaced font and asked nothing of you.

1. Focus mode dims everything except the current sentence or paragraph. The writer's working memory holds about one sentence; the rest of the page is a distraction that invites rereading and polishing too early. Cost: you lose peripheral context, so it is a mode you turn on, not the default.
2. A typeface designed for the app. Early versions used a customized version of Nitti, a monospace by Bold Monday. Around 2017 and 2018 iA made its own family, iA Writer Mono, Duo and Quattro, derived from IBM Plex Mono and released for free. Duo is "duospaced" (most letters share one width, wide letters like m and w get more) and Quattro uses four widths, keeping a typewriter's draft feel while reading more evenly. Cost: the look is unmistakable, and people who want a proportional serif must pick from very few choices.
3. Markdown marks stay visible, but quiet. The # of a heading sits grey in the margin, so the text stays plain and portable while structure is still legible. Cost: newcomers must learn a little syntax; there is no bold button doing the work for them.
4. The library is your file system. Documents are plain text files in folders you own (iCloud, Dropbox, a local folder). No proprietary database, no lock-in. Cost: no rich metadata, and sync problems become the file provider's problems, which users still blame on the app.
5. Content blocks. Typing a path like /chart.png on its own line pulls in an image, a CSV table or another text file. It keeps the document text-only while allowing rich output. Cost: it is a power feature that few users discover without docs.
6. Authorship (iA Writer 7, late 2023). Text you paste from an AI tool can be marked as not yours and shown in a distinct style until you rewrite it; you can also attribute text to named authors. The app takes a public position: AI as a sparring partner, not a ghostwriter. Cost: extra friction for people who want the machine to write for them, which is the point.
7. Tools that show, not change. Syntax highlighting can colour adjectives, nouns, adverbs, verbs and conjunctions, and Style Check flags filler words and clichés. They hold a mirror to your prose instead of rewriting it. Cost: they are opinionated and English-first.
8. Very few settings, on purpose. iA has argued for years that each preference is a decision the designer pushed onto the user. The app has grown some settings over time, but each is resisted. Cost: you cannot tune it to taste, and some people leave for that reason. This is "the no-settings price".

Trade-offs and criticisms

The usual criticisms are fair and worth learning from. Monospace is not the best face for long reading, so iA added Duo and Quattro and a preview mode in proportional type. The lack of settings frustrates writers with established habits. The file-based library means features that need a database (tags across files, backlinks) arrive late or not at all, because each one has to work on plain files. And opinionated tools like Style Check can feel preachy. Each criticism is the shadow of a decision that also makes the app good.

Related products

• UlyssesThe Soulmen, 2003; iOS later
  A Markdown-style writing app with a managed library instead of plain folders, which enables goals, sheets and publishing. Same calm, opposite bet on storage.
• Typora2015
  Hides Markdown marks as you type and renders inline. The other answer to "should syntax be visible?"
• WriteRoomHog Bay Software, 2006
  An early full-screen "distraction-free" Mac editor. Focus as a product idea predates iA; iA made it typographic.

Things 3 shipped in 2017 after years of quiet work, and won an Apple Design Award. Its competitors had more features: shared lists, comments, integrations. Things had a better model of a single person's day. It still sells as separate one-time purchases per platform, with free sync, and no subscription.

1. Today is a list you curate, not a filter of due dates. You decide what goes in Today. Due dates and scheduled dates feed it, but the list is a promise to yourself, short enough to finish. Cost: people who want an automatic "everything due" view must learn the difference between "when" and "deadline".
2. This Evening. A tiny split that matches how many people live: work things, then home things. It keeps Today from feeling like a single wall. Cost: almost none, which is why it is a lovely decision.
3. The Magic Plus button. Tap it to add at the end; drag it to place a new to-do exactly where you want it in the list, or drop it on the left to send to Inbox, or create a heading. One control, many outcomes, all learned by direct manipulation. Cost: discoverability. People who only tap never find the drag.
4. A small, fixed set of lists. Inbox, Today, Upcoming, Anytime, Someday, Logbook. Not a user-defined system. The model comes from the Getting Things Done tradition, simplified. Cost: people with an elaborate personal system feel boxed in.
5. Headings inside projects. A project can be divided into sections without becoming sub-projects. Shallow structure stays shallow. Cost: no deep nesting, by design.
6. Deadlines are rare and red. A deadline is separate from "when" and shown with a quiet countdown. Red appears only there, so it means something. Cost: you need to understand two date concepts.
7. Tags are light and optional. Small grey lozenges, filterable, never required. Cost: tag-heavy users want more.
8. Quick Entry everywhere, with natural dates. On the Mac a global shortcut opens a small window; you can type "tomorrow" or "next fri" in the when field. Capture must cost under five seconds or ideas are lost. Cost: another shortcut to learn.
9. No collaboration, by design. No shared lists, no assignees, no comments. Cost: families and teams leave for Todoist, Reminders or a team tool. Benefit: none of the screens carry the weight of other people.

Trade-offs and criticisms

Things releases slowly; years can pass between visible features, and some users read that as neglect. It is Apple-only, so anyone on Android or Windows is out. Its refusal of collaboration rules out households. And some power users find the fixed model too gentle: no task dependencies, limited recurrence logic compared with OmniFocus. Every criticism points to a door Cultured Code chose to keep closed. The lesson is that saying no costs customers, and the product is better for those who stay.

Related products

• OmniFocusThe Omni Group, 2008
  A deep Getting Things Done tool with perspectives, contexts and review. The opposite bet: power over calm. Both are coherent.
• TodoistDoist, 2007
  Cross-platform, natural-language entry ("every Monday at 9"), shared projects. Wins on reach and sharing where Things refuses.
• Apple RemindersApple, 2011
  Free, shared lists, grocery sections. The default you are always competing with.

Notes apps live or die on the first ten seconds: an idea arrives, and the tool must catch it before it fades. Apple Notes takes the shortest path: the first line you type becomes the title, there is no "save", and the note lives in the last folder you used. Bear, which won an Apple Design Award in 2017, keeps that speed and adds a structure most people can learn in one sentence: type #work/clients anywhere in a note, and the note appears under that nested tag in the sidebar.

1. Tags are the folders. Bear has no separate folder system. Nested tags (#work/clients) give hierarchy, and a note can live in several places at once. Cost: a folder-minded person has to unlearn the idea that a note lives in one place.
2. Structure is typed, not filed. You never stop writing to organize. Typing the tag is the filing. Cost: renaming a tag means rewriting text in many notes (Bear does this for you, which is a necessary feature that comes with the choice).
3. Typography is a feature. Bear ships carefully chosen themes and fonts, generous line spacing and a reading width. Notes feel like something worth rereading. Cost: design effort that does not show up in a feature list.
4. Wiki-style links. Double-bracket links connect notes by title, a lightweight version of what Obsidian does with files. Cost: links depend on titles staying stable.
5. One search box does most of the work. Special search terms exist, but plain search is the main tool. Cost: power filtering is less visible.
6. The Panda editor (Bear 2, 2023). Bear rebuilt its editor to be closer to standard Markdown, so text moves more cleanly in and out. Cost: a rewrite of the core of the product, years of work for a small team, with changes that some long-time users had to adjust to.
7. Apple Notes: zero decisions at capture. No title field, no save, no folder prompt. The first line is the title. Cost: notes pile up unsorted; Apple added tags, smart folders and Quick Note later to help.
8. Apple Notes: the free default. Pre-installed, synced, shared. Its existence sets the bar every paid notes app must clear. Cost to competitors, not to Apple.

Trade-offs and criticisms

Bear is Apple-only for its full app (a web version arrived later), and its sync relies on iCloud, which limits who can use it. Tags-as-folders confuses people who want one canonical location. Apple Notes, for its part, has no export story most people trust, uses a closed format, and gets messy at scale. Both are good at the first ten seconds and less good at the ten-thousandth note.

Related products

• SimplenoteAutomattic, 2008 onward
  Plain text, tags, fast sync, free. The minimum viable notes app, still useful.
• Google KeepGoogle, 2013
  Card grid with colours and labels. Capture-first on Android, with visual scanning instead of typography.
• Craft2020
  Blocks and rich pages with a native feel. A middle road between Bear and Notion.

A blank journal is intimidating. Day One's design answer is that an entry should never start fully blank. It already knows the date, often the place and the weather, and it can suggest photos you took today. And it offers the strongest reason to keep journaling: "On This Day", which shows what you wrote a year or five years ago. The reward for writing today is a future surprise. Apple's own Journal app (iOS 17.2, December 2023) took a similar path with "journaling suggestions" drawn from your photos, places and workouts.

1. Automatic metadata. Date, location, weather and device are attached without asking. An entry is never empty, and old entries are rich with context you would never type. Cost: privacy questions, which is why location can be turned off and why encryption matters.
2. Photos as a starting point. A picture from today is an easy first line. Cost: a photo-heavy journal can become a photo album without reflection.
3. On This Day. The return loop is your own past, not a streak or a points system. It rewards consistency with meaning rather than numbers. Cost: it can surface painful memories; good apps let you control or hide it.
4. Prompts and templates. For the "I don't know what to write" moment. Cost: prompts can make entries feel like homework if pushed too often.
5. End-to-end encryption. Journals are among the most private data a person creates. Day One added end-to-end encryption for synced journals, turning privacy from a policy into a property. Cost: if you lose your key, Day One cannot recover your data. The design must explain this without scaring people.
6. Multiple journals with colours. Work, family, travel, dreams. Separation lets people write honestly in each. Cost: one more decision at capture, softened by a default journal.
7. Gentle reminders. A daily reminder at a time you choose, with no guilt copy. Compare Duolingo in chapter 19. Cost: lower engagement numbers than a nagging app, probably.

Trade-offs and criticisms

Day One's richest features sit behind a Premium subscription, and long-time users of the original paid app did not all welcome the change. The Automattic acquisition reassured some (a stable owner) and worried others. Apple Journal, being free and built in, takes the casual journaler; Day One must win on depth, export, cross-platform reach and longevity. And automatic metadata only helps if people trust it: one creepy location tag and the whole system feels like surveillance.

Related products

• Apple JournalApple, iOS 17.2, 2023
  Suggestions from your photos, workouts and places, chosen on device. The platform owner can do automatic context with less creepiness because data stays on the phone.
• Google Photos "Memories"Google
  Resurfaces photos from past years. The same return loop without writing.
• Stoic, Reflectly and others2010s–2020s
  Prompt-led journaling with mood tracking. More structure, more guidance, less freedom.

Before Linear, issue trackers were mostly configurable databases: custom fields, custom workflows, slow pages, many clicks. Linear's founders came from design and engineering roles at large tech companies and built the tracker they wanted. The product feels instant because the data your screen needs is already on your machine. Changes appear immediately and sync in the background. That architectural choice is a design choice: it is the only way to make every click feel like nothing happened between intention and result.

1. A command menu for everything (Cmd+K). Every action is reachable by typing its name. A newcomer can search; an expert can type. Cost: the menu must be maintained as carefully as the visible UI, and actions need good names.
2. Shortcuts are shown where actions are listed. The menu teaches its own shortcuts: after choosing "Assign to…" a few times you notice the A beside it and start pressing A. Cost: shortcut collisions get harder as features grow.
3. Instant by architecture. Linear keeps workspace data in the client and applies changes locally first (optimistic updates), syncing in the background. Every action feels like it happened when you pressed the key. Cost: a complex sync engine and conflict handling, a large engineering investment for a "feeling".
4. Cycles, not open-ended sprints. Teams work in fixed time boxes that start and end automatically, and unfinished work rolls over. Planning becomes a rhythm rather than a meeting. Cost: teams with other rhythms must bend to it.
5. Triage as a default inbox. New issues from other teams or integrations land in Triage, so someone decides before anything enters the backlog. Cost: someone must own triage.
6. Dense, typographic rows. Small status shapes, monospaced IDs, priority bars instead of words. Fifty issues fit on a laptop screen. Cost: newcomers need a moment to learn the symbols.
7. Opinions, published. "The Linear Method" is a public set of practices (for example, short specs, momentum over ceremony). The product implements the opinions; the writing explains them. Cost: customers who disagree feel the tool fighting them.
8. Craft as marketing. Linear's website and release notes are carefully made, and its dark, gradient-lit landing-page style was copied widely across the industry in the early 2020s. Cost: imitators made the style generic, which pushes Linear to keep evolving.

Trade-offs and criticisms

Opinionated workflow is a selling point until your team works differently; then it is friction. Dense, symbol-heavy rows are fast for daily users and confusing for occasional ones (designers or managers who visit once a week). And the "Linear look" became so common in startup marketing that it now signals "generic SaaS" as often as it signals craft. The deeper lesson survives the fashion: the speed is architectural, and you cannot copy it with a dark theme.

Related products

• JiraAtlassian, 2002
  Configurable to any process, used by huge organizations. Flexibility serves big companies and slows individuals; Linear chose the opposite end.
• GitHub IssuesGitHub
  Lightweight, close to the code, keyboard shortcuts and a command palette. Proximity to work can beat a separate tool.
• Height, Plane and others2020s
  Trackers that adopted command menus and fast UIs. A good pattern becomes the baseline quickly.

Email is a solved problem in the sense that Gmail is free and works. Superhuman asked a very high monthly price (around $30 a month for years) and found people who paid it gladly. The design explains why: for someone who spends hours a day in email, saving minutes per hour is worth real money. Every decision serves that one customer, and the company was unusually explicit about choosing that customer, using a survey method that has since become standard reading for founders.

1. The 100 ms rule. The team set a target that interactions should respond within about 100 milliseconds, the threshold at which people perceive a response as instant. Speed is not polished later; it is a requirement on every feature. Cost: features that cannot be fast are delayed or refused.
2. Keyboard for everything, taught deliberately. Every action has a shortcut, and the command menu shows them. Hints nudge you when you use the mouse for something that has a key. Cost: a learning curve the product must pay for with teaching.
3. One-to-one onboarding. For years, new users got a live session with a Superhuman specialist who set up their inbox and drilled shortcuts. It is expensive and does not scale, which was part of the point: it guaranteed that early users reached the fast state, and it taught the company what confused them. Cost: slow growth; later the company shifted toward self-serve onboarding.
4. Split inbox. Mail is divided into streams (important people, team, news, calendar and so on), so you triage one kind of mail at a time. Cost: some setup, and mail occasionally lands in the wrong split.
5. Opinions about inbox zero. "Done" (archive) is the main action, reminders bring mail back later, and an empty inbox is celebrated with calm imagery. Cost: people who use email as an archive or a to-do list may find the push uncomfortable.
6. A premium price as a filter. A high price signals seriousness and funds the onboarding and speed work. Cost: most people will never try it.
7. Read statuses, and a correction. Superhuman originally showed whether and where a recipient opened your email, using tracking pixels, on by default. After public criticism in 2019 (notably a widely read post by designer Mike Davidson), it removed location data and turned read statuses off by default. Cost: a lesson in how a "power feature" for one user can be surveillance for another.

The product-market fit survey

In a 2018 First Round Review essay, Vohra described how Superhuman used a single question popularized by Sean Ellis: "How would you feel if you could no longer use this product?" with answers very disappointed, somewhat disappointed, not disappointed. Ellis's benchmark is that products with fit tend to have at least 40% "very disappointed". Superhuman started at about 22%. It then segmented respondents, focused on the people who loved it, studied what they valued, asked the "somewhat disappointed" what held them back, and spent its roadmap roughly half on doubling down and half on removing those blockers. Vohra reported the score rising to 58% over the following quarters.

This matters to a designer because it turns "is this good?" into a number you can watch and a segment you can design for. Play with the calculator below.

Trade-offs and criticisms

The price excludes most people, and early invite-only access read to some as exclusivity theatre. One-to-one onboarding does not scale, and a product that needs a human to teach it has a design debt to repay eventually. The read-receipt episode showed how a feature optimized for one user (the sender) can harm another (the recipient). And the core insight, speed, is copyable: Gmail, Spark and others added split inboxes and better shortcuts. What remains distinctive is the discipline of saying no to anything slow.

Related products

• GmailGoogle, 2004
  Free, keyboard shortcuts (opt-in), tabs for categories. The default; Superhuman's design only matters where Gmail is slow.
• HEY37signals, 2020
  A different opinion: screen first-time senders, block tracking pixels, separate "Imbox" from "The Feed" and "Paper Trail". Opinionated email that chose privacy as its edge.
• SparkReaddle
  A smart inbox that groups mail by type, free tier plus team features. The split-inbox idea at a mass-market price.

Menus scale badly. A menu with eight items is easy; with eighty it is a maze. A launcher replaces navigation with recall: you know the name of the thing you want, so type it. Spotlight brought the idea to the Mac in 2005 (Command-Space). Quicksilver and LaunchBar had explored it earlier, Alfred built a business on it with paid workflows, and Raycast turned it into a developer platform where anyone can write a command in TypeScript and React that renders with Raycast's own list, form and detail components.

1. One global shortcut, one field. The entire product is behind one key combination. There is no window to manage. Cost: if users never learn the shortcut, the product does not exist for them.
2. Fuzzy matching and learned ranking. "cl" finds Clipboard History, Calendar and "Create Linear issue". Results you choose often rise to the top. Cost: rankings shift as habits change, which occasionally surprises muscle memory.
3. Everything is a command in one list. Apps, files, system actions, calculations and extension commands share one result list. Cost: name collisions, solved by labelling the type of each result.
4. Primary action plus an actions menu. Enter does the obvious thing; Command-K shows every other action for the selected item. Two levels, consistent everywhere. Cost: actions must be named and ordered with care for every result type.
5. Extensions use the host's components. Raycast extensions are written with its UI components (lists, forms, details), so every third-party command looks native and keyboard-complete. Cost: extension authors give up visual freedom; that is the point.
6. Built-in utilities that replace small apps. Clipboard history, window management, snippets and quicklinks. Cost: Raycast competes with many small developers' products.
7. Business model: free core, paid Pro and team features. Raycast's core is free; paid plans add AI features, sync and team sharing. Alfred sells a one-time Powerpack license for workflows. Spotlight is part of the OS. Each model shapes how much the launcher pushes you toward its extras.

Trade-offs and criticisms

Launchers are invisible to people who do not know they exist, so they serve experts first. Fuzzy ranking that learns can feel unpredictable. Raycast's breadth (AI chat, notes, window management, clipboard) risks becoming the cluttered menu it set out to replace, and its extension store means quality varies between authors. Spotlight's newer versions (Apple expanded it substantially with actions in macOS 26, 2025) narrowed the gap for casual users, which is the usual fate of a third-party pattern the platform owner decides to absorb.

Related products

• QuicksilverBlacktree, 2003
  Noun-verb commands: pick a thing, then an action. The ancestor of Raycast's action menu.
• VS Code Command PaletteMicrosoft, 2015
  Ctrl/Cmd-Shift-P for every editor command. Made the palette standard in developer tools.
• Slack, Notion, Figma quick actions2010s–2020s
  Cmd-K or "/" menus inside apps. The launcher pattern moved inside products.

In 2010, accepting card payments online meant merchant accounts, gateways, forms and weeks of paperwork. Stripe's early pitch was that a developer could start accepting payments with a few lines of code. That promise moved the design work to unusual places: the docs page, the API's naming, the JSON of an error response, the test card number 4242 4242 4242 4242. For a developer, those are the interface.

1. Three-column docs: navigate, read, copy. Prose and the code it describes sit side by side, so you never scroll between explanation and example. Cost: on narrow screens the layout must collapse carefully, and every prose change needs a matching code change.
2. Personalized samples. When you are signed in, code samples can include your own test API key, so pasted code runs as is. Cost: docs become a logged-in product with security considerations (only test keys, never live ones).
3. Test mode as a first-class world. Separate test keys (sk_test_…), a dashboard toggle and documented test cards for success, decline and authentication. Developers can rehearse every path, including failure. Cost: two parallel environments to keep consistent.
4. Errors written for the reader. An API error carries a machine type, a code, the parameter at fault, a human message and often a link to the docs. Errors are documentation that arrives exactly when needed. Cost: every error path needs writing and maintenance.
5. Consistent, guessable naming. Objects and parameters follow strict conventions, and API versions are pinned per account so changes do not break integrations. Cost: heavy internal review for every new endpoint.
6. Checkout: from a modal to a hosted page. Early Stripe Checkout was a pop-up form on the merchant's site. In 2019 Stripe launched a new hosted Checkout page, which made it easier to support new rules (such as Europe's Strong Customer Authentication), wallets and local payment methods without each merchant rebuilding forms. Cost: merchants give up some control over their page.
7. A brand that signals care. Stripe's site is known for its animated gradients and precise typography. For a company selling invisible infrastructure, visible care is evidence. Cost: high expectations for every page, and the gradient became an industry cliché.

An error message as interface

Compare the two responses below. Both report the same problem. Only one lets a tired developer fix it in under a minute.

HTTP 400
{ "error": "Bad Request" }

HTTP 402
{ "error": {
  "type": "card_error",
  "code": "card_declined",
  "decline_code": "insufficient_funds",
  "message": "Your card has insufficient funds.",
  "param": "payment_method",
  "doc_url": "https://…/error-codes#card-declined"
} }

Trade-offs and criticisms

Stripe's product surface has grown enormous (billing, tax, identity, issuing, terminals and more), and its docs and dashboard now carry that complexity; newcomers can feel lost where they once felt welcomed. Fees are higher than some local alternatives, so in markets like India many businesses choose local gateways. Hosted Checkout trades merchant control for compliance and conversion. And the visual brand's success turned "Stripe gradient" into a template that many startups copy without the substance behind it.

Related products

• Twilio2008
  Developer-first communications API with quick-start docs. DX as a go-to-market strategy predates and parallels Stripe.
• RazorpayIndia, 2014
  Developer-friendly payments for Indian businesses, with UPI and local methods built in. The same DX lesson adapted to local rails.
• Vercel, Supabase2020s
  Docs with copyable commands, clear error pages and dashboards built around developer flow. A whole generation of tools absorbed the Stripe approach.

Apple's Camera app is built for everyone and hides almost every control. Pro camera apps before Halide often went the other way, exposing every setting as a tiny button. Halide's founders (a designer and an engineer, both photographers) asked what a physical camera does well: controls are where your fingers already are, and each control has one job. Halide maps that onto glass. It also writes publicly and in depth about iPhone camera hardware on its blog, which educates its users and markets the app at once.

1. Gestures on the image, not buttons over it. Swiping vertically anywhere on the viewfinder adjusts exposure. The whole screen is the control, so you keep your eye on the picture. Cost: gestures must be taught, and accidental swipes must be easy to undo (a tap returns to auto).
2. Manual focus as a dial under the thumb. Focus is adjusted by sliding along a control near the bottom corner, the way a lens ring sits under your fingers. Cost: left-handed use and large phones need care.
3. Focus peaking. While focusing manually, edges that are in focus are outlined in colour. It borrows a decades-old aid from video and mirrorless cameras to answer the question "is it sharp?" on a small screen. Cost: visual noise when on, so it appears only during manual focus.
4. Live histogram and readouts. Small, legible, honest data about the exposure. Pros read them; beginners can ignore them. Cost: screen space.
5. Reachability first. Frequently touched controls cluster at the bottom, within a one-handed thumb arc. Rarely touched indicators sit at the top. Cost: the top of the screen becomes "read only", which some users find odd at first.
6. Opinionated output options. Halide offers RAW capture and, from 2024, "Process Zero", a mode that skips Apple's heavy computational processing for a more natural, film-like file. The product takes a position on what a photo should be. Cost: images look less "finished" than the default camera's, which some people dislike.
7. Teaching as marketing. Long blog posts analyse each new iPhone's camera in detail. Users learn the craft, and trust the app made by people who know it. Cost: time spent writing instead of building, which is a fine trade for a small team.

Trade-offs and criticisms

Gesture-heavy interfaces are invisible until learned, and Halide relies on a short tutorial and on users who want to learn. Pro features shift with every iPhone generation, so the app must constantly chase hardware changes. Some photographers want more manual controls, others fewer, and the app's personality (dark, precise, camera-like) suits some tastes more than others. The pricing, a subscription with a one-time purchase option, tries to serve both the casual and the committed.

Related products

• Apple CameraApple
  Minimal controls, heavy automatic processing. The default Halide defines itself against.
• VSCO2012
  Film-like presets and a minimal camera. Taste as product, aimed at mood rather than control.
• Lightroom Mobile, ProCamera, Blackmagic Camera2010s–2020s
  Manual control apps with different densities. Compare how each solves reach and visibility.

Marco Arment built Instapaper and was an early engineer at Tumblr before making Overcast alone. Podcast apps mostly compete on directories and lists. Overcast competes on what you hear. Smart Speed shortens silences in speech without changing the speed of voices, and the app tells you how much time it has saved you in total. Voice Boost evens out loudness and adds clarity so quiet hosts and loud guests sound balanced, which matters in a car or on a noisy bus.

1. Smart Speed: remove silence, not syllables. Shortening pauses saves time without making voices sound sped up. It is a feature you cannot see, only hear. Cost: occasional clipped dramatic pauses, and heavy engineering in audio processing.
2. Voice Boost: fix the audio, not the UI. Loudness normalization and clarity, tuned for speech. Cost: it alters the producer's mix, which some audio purists dislike, so it is a toggle.
3. Show the invisible benefit. A cumulative "time saved" figure turns a hidden feature into a felt one. Cost: almost none, and it is a quiet form of marketing users share.
4. Big controls for distracted moments. Skip buttons and play are large because podcasts are heard while walking, cooking or driving. Cost: less room for metadata.
5. Business model in public. Overcast began with a paid unlock, moved in 2015 to all features free with optional patronage, and later settled on unobtrusive ads plus a paid subscription that removes them. Arment explained each change publicly, including his reasoning about not tracking listeners. Cost: lower revenue than an aggressive model; gain: trust.
6. Scope set by one person's capacity. iPhone, iPad and Watch, with a web player; no Android app. One developer cannot do everything, so the platform choice is a decision about quality. Cost: a large market not served.
7. Redesign when the foundation ages. Arment has rewritten large parts of the app over the years, including a major rewrite and redesign in 2024, accepting the cost of change to keep a one-person codebase maintainable. Cost: some long-time users disliked the new layout at first.

Trade-offs and criticisms

A one-person product moves at one person's pace: feature requests wait, and some never come (no Android). Opinionated redesigns can upset long-time users, and with no team to absorb feedback the developer carries all of it. The ad-supported tier depends on the podcast ad market. And because the best features are audio, they are hard to show in screenshots and App Store pages, which is why "time saved" matters so much.

Related products

• Pocket CastsShifty Jelly, 2010; now Automattic
  Cross-platform, "Trim Silence" and volume boost. The same audio ideas with broader platform reach.
• Castro2014
  A triage model: new episodes go to an inbox, you choose what enters the queue. A different signature: workflow rather than audio.
• Apple PodcastsApple
  The pre-installed default. Third-party players must be distinctly better at something specific.

You met Silvio Rizzi in the starter guide. This breakdown looks at a rare thing: a successful product that changed its core model. Classic Reeder treated feeds like email: every article is unread until you read it or mark it read, and the counts pile up. Many people experience that as a chore. In 2024 Rizzi released a new app called simply Reeder, which pulls in RSS, YouTube, Reddit, Mastodon, Bluesky, podcasts and more into one timeline, remembers your position across devices, and drops unread counts entirely. The previous app continues as Reeder Classic.

1. Unread counts removed. A number on every feed turns reading into a debt. The new Reeder replaces "what have I not read?" with "where was I?". Cost: completionists who liked clearing their inbox felt the product had been taken away; hence Reeder Classic remains available.
2. One timeline for many sources. Articles, videos, social posts and podcasts arrive in the same chronological stream, each with a light source label. Cost: mixing media types makes typography and layout harder, and some users want separation.
3. Position sync as the core promise. Pick up exactly where you left off on any device. It is what makes a timeline without counts work at all. Cost: sync must be reliable; a lost position breaks the model.
4. Reading typography first. Both apps open articles in a clean reader view with careful line length, spacing and fonts. Cost: original site styling is lost, which is the point.
5. Gestures carried over. Swipes for common actions (save, star, share) and pull gestures keep the feel that made Reeder famous. Cost: discoverability, as always with gestures.
6. Keeping the old app alive. Instead of forcing the change, Rizzi kept Reeder Classic for people who want the inbox model. Cost: two products to maintain for a solo developer.
7. A new business model with the new app. The new Reeder is sold as a subscription with a trial, where earlier versions were one-time purchases. Cost: some loyal users resisted the change; gain: income for continuous work on many integrations.

Trade-offs and criticisms

The rethink split the audience, as any model change does. Reviewers noted that the timeline is great for skimming but worse for "I must read every post from this one blog". Some missed the feature of marking things read. The lesson is not that the timeline is right. It is that a mental model (inbox versus stream) is the deepest design decision in a product, and changing it is a new product, which is why Rizzi shipped it as one rather than as Reeder 6.

Related products

• NetNewsWireBrent Simmons, 2002; open source since 2019
  A free, native, inbox-model reader. The classic model done with care and no business model at all.
• Feedbin, Inoreader, FeedlyWeb services
  Sync services and readers with different densities and AI features. The backend market Reeder relies on.
• Readwise Reader2022
  Feeds plus read-later plus highlighting. See Breakdowns: Reading.

Tweetie was a Twitter client for iPhone that was fast, dense and elegant at a time when most apps were none of these. Tweetie 2 introduced pull to refresh: drag the timeline down past the top, an arrow flips to say "release", let go, and new tweets load. Twitter bought atebits in 2010 and Tweetie became the official Twitter app. Brichter also wrote openly about how he made scrolling fast by drawing each table cell as a single flat image rather than many layered views, a reminder that "feel" is often an engineering decision. In 2012 he released Letterpress, a two-player word game whose flat, bold colour fields looked strikingly different from the textured iOS design of the time, a year before iOS 7 went flat.

1. Pull to refresh uses a motion users already make. People scroll to the top to see what is new. Extending that motion past the edge needs no new button and no new concept. Cost: the action is invisible until learned (the text hint solves it).
2. A visible threshold. The arrow flips and the label changes from "pull" to "release" at the commit point, so you can cancel by not pulling far enough. Feedback makes a gesture safe. Cost: none worth naming; it is a model of good gesture design.
3. New content arrives where you are looking. Fresh items appear at the top of the list, under your thumb. Cost: the list position must be managed so you do not lose your place.
4. Performance as design. Fast, smooth scrolling made the dense timeline readable in motion. On 2008 iPhone hardware, that required custom drawing. Cost: a lot of low-level engineering for something users only notice when it is missing.
5. Letterpress: a flat colour language. Tiles take your colour or your opponent's; protected tiles become a darker shade. The whole game state is readable at a glance from colour alone. Cost: colour must be paired with shade differences for colour-blind players (Letterpress offered alternative colour themes).
6. Letterpress: turns, not real time. Built on Game Center's turn-based play, games can last days. It is a calm social game, the opposite of an arcade. Cost: games can stall when opponents vanish.
7. Letterpress: generous free, simple paid. The game was free to play, with a one-time purchase unlocking extra features such as unlimited simultaneous games and themes. Cost: lower revenue than ads or energy timers; gain: a clean experience.

Trade-offs and criticisms

Pull to refresh became so common that it spread into apps where it made no sense (lists that update live anyway), and Brichter himself has said in interviews that he has mixed feelings about its overuse and about the pull-to-refresh "slot machine" feeling in social feeds. Twitter later held a patent on the gesture, though it committed to using it only defensively. Letterpress's flat colour language, admired in 2012, helped push a trend that sometimes went too far, stripping affordances from buttons across the industry. Inventions escape their context; the designer cannot control what the industry does with them.

Related products

• iOS and Android lists2013 onward
  Platform refresh controls (UIRefreshControl on iOS, SwipeRefreshLayout on Android) made the gesture standard. A third-party invention became a platform convention.
• ClearRealmac and Impending, 2012
  A to-do app built almost entirely on gestures (pinch, pull, swipe). Shows the limit: delightful, but hard to learn.
• WordleJosh Wardle, 2021
  A word game with colour as state and one puzzle a day. Letterpress's calm, readable-state lesson in a new form.

In Notion, a paragraph, a heading, an image, a to-do and an entire database are all blocks. Type / and you can turn any line into any block. Pages contain blocks; databases contain pages; views show databases as tables, boards, calendars or galleries. With a few primitives you can build a wiki, a CRM or a habit tracker. In Obsidian, a note is a .md file in a folder (a "vault"). You connect notes with [[double bracket links]], and the app shows backlinks and a graph of connections. Obsidian's CEO, Steph Ango, summed up the philosophy in a 2023 essay titled "File over app": the files should outlive the app.

Flexibility versus opinion

Notion and Obsidian are both flexible, and both pay the flexibility tax: the user becomes the designer of their own system. That is a joy for some and a burden for most. Compare Things (chapter 4) and Bear (chapter 5), which make the system decisions for you. Neither approach is right in general. Ask who your user is. A person who enjoys building systems will love blocks and plugins. A person who wants to get on with work will abandon them after the first weekend.

Related products

• Roam Research2019
  Popularized bidirectional links and daily notes for a devoted audience. Obsidian adopted the ideas with local files.
• Coda, Airtable2010s
  Docs-as-apps and spreadsheet-databases. Notion's database side, pushed further toward building tools.
• Logseq2020
  Open-source outliner on local files. Blocks and files combined, with different trade-offs.

Before Figma, interface design happened in desktop apps (Sketch was the favourite), and sharing meant exporting, syncing files and naming versions like final_v7_REAL.sketch. Figma rendered a full vector design tool in the browser using WebGL and compiled C++ (later WebAssembly), which looked like an engineering stunt. The design consequence was the real point: a design became a URL. Anyone with the link could look, comment or edit, at the same time, with no install.

1. The browser as the platform. No install, any operating system, always the latest version. A design is a link. Cost: a huge engineering bet on rendering performance in the browser, and dependence on connectivity (desktop apps exist, but they wrap the same web engine).
2. Multiplayer by default. Live cursors with names show who is where. You can click an avatar to follow someone's view during a review. Cost: the anxiety of being watched while you work, and conflicts that need careful merging on the server.
3. Comments pinned to the canvas. Feedback lives on the exact pixel it is about, not in an email thread. Cost: comment clutter on busy files.
4. Constraints, components and auto layout. Frames resize like real interfaces; components and variants enforce consistency; auto layout (2019) brought flexbox-like behaviour. Design files moved closer to code. Cost: a steeper learning curve than drawing rectangles.
5. Free viewers, paid editors. Anyone can view and comment, so a designer pulls the whole company into the tool. Growth came from sharing. Cost: pricing tension as companies try to minimize paid seats.
6. Community and plugins. Public files, templates and plugins (from 2019) turned users into contributors. Cost: moderation and quality control.
7. Expansion beyond design. FigJam (2021) for whiteboarding, Dev Mode (2023) for developers, and later AI and site-building features. Cost: a growing, more complex product; an attempted acquisition by Adobe, announced in 2022, was abandoned in late 2023 after regulatory scrutiny.

Trade-offs and criticisms

Figma made collaboration so easy that it changed design culture, not always for the better: work in progress is visible to everyone, and some designers describe feeling watched, or note that reviews multiply because they are so easy. The product has grown heavy as it expanded. Pricing changes have frustrated teams. And because Figma became the standard, alternatives struggle, which concentrates power over a profession's main tool in one company.

Related products

• Google Docs2006
  Real-time co-editing of documents in the browser. Figma brought the same model to a far harder medium.
• SketchBohemian Coding, 2010
  The Mac-native tool that defined UI design in the 2010s, later adding collaboration. A native-quality single-user tool overtaken by a collaborative model.
• PenpotOpen source
  An open-source, browser-based design tool built on web standards. An answer to concentration of power in one vendor.

Browsers had barely changed in a decade: a row of tabs that shrink until unreadable. Arc moved tabs into a vertical sidebar, separated work and personal browsing into Spaces with their own colours, split tabs into pinned (kept) and today (temporary), and archived temporary tabs automatically after a period of inactivity (12 hours by default). It was beautifully made and deeply loved by a minority. In October 2024 the company told users it would stop major feature development on Arc to focus on a new product, and in 2025 it released Dia, a simpler browser built around AI. In a public letter, CEO Josh Miller reflected that Arc had been too different, too many new things to learn, for most people. In September 2025 Atlassian announced it would acquire the company.

1. Vertical tabs in a sidebar. Titles stay readable however many tabs you open. Cost: horizontal space, and a break with twenty years of habit.
2. Pinned versus today. Two kinds of tab for two intentions: things you return to, and things you are just looking at. Cost: a new concept to learn on day one.
3. Auto-archive. Unpinned tabs you have not viewed for a while (12 hours by default, adjustable) slide into an archive. Tab hoarding is solved by forgetting on your behalf. Cost: the occasional panic of "where did that tab go?", softened by a searchable archive.
4. Spaces with their own colour and profile. Work and home are separated visually and by login. Cost: more structure to set up.
5. The browser as a creative object. Boosts (restyle any site), Easels (collage boards), split view and Little Arc windows. Cost: feature sprawl; each delight is another thing to learn.
6. Strong visual identity. Coloured gradients, playful motion and a carefully crafted onboarding. Cost: high expectations and a look that ages quickly.
7. The decision to stop. The team concluded that Arc's learning curve limited it to enthusiasts, and redirected effort to a simpler product. Cost: disappointed loyal users. Gain: a lesson, stated in public, that most products never admit.

Trade-offs and criticisms

Arc is the clearest case in this guide of the novelty tax: every new concept (Spaces, pinned versus today, archive, Boosts) was good on its own, and together they were too much for people who just wanted to open a website. Fans loved it precisely because of that richness. The business never found a revenue model before the pivot. And the turn to an AI browser raised questions about whether the company's attention had followed the product or the hype. Whatever you think of the pivot, the honest public reflection is rare and worth reading.

Related products

• Safari and Edge tab groups, vertical tabs2020s
  Mainstream browsers adopted vertical tabs and groups as options. Bold products often seed ideas that the incumbents adopt gently.
• Zen BrowserOpen source, Firefox-based
  Brought Arc-like sidebars and workspaces to Firefox. A loved model outlives its product.
• DiaThe Browser Company, 2025
  A conventional-looking browser with an AI assistant built in. The same team chose familiarity first the second time.

In the late 1990s, rival search sites became "portals" stuffed with news, weather, stock prices and ads. Google's homepage offered a logo, a box and two buttons. The emptiness made the job obvious and the page fast on dial-up modems. Craigslist, meanwhile, looks like a page from 1999 because, in spirit, it is one: a city, a grid of category links, then lists of text listings. Designers love to mock it, and it has served huge numbers of people for decades. Both pages teach the same lesson from opposite ends: the right amount of design depends on the job.

1. Google: one field, centred. The page has one job and shows one control. Cost: Google had to resist years of pressure to use the most visited page on the web for promotion.
2. Google: the results page carries the density. The homepage is empty because the results page is full: titles, links and snippets you scan in seconds. Simplicity at the entry, density where the work happens. Cost: as results pages filled with ads, cards and, from 2024, AI-generated answers, the clarity of the ten blue links eroded.
3. Google: play, carefully placed. Doodles change the logo for holidays and anniversaries without touching the box. Personality lives in one controlled spot. Cost: almost none.
4. Google: speed as a value. Results arrive fast and the page is light. Every extra kilobyte was treated as a cost. Cost: constant engineering discipline.
5. Craigslist: everything on one screen. The home page lists every category as a link, so you navigate by scanning, not clicking through menus. Cost: overwhelming at first glance; rewarding after one visit.
6. Craigslist: text over images. Listings are rows of text with price and place, so you scan dozens per screen. Images appear inside listings. Cost: less appealing for visual goods; competitors with photo grids took some categories.
7. Craigslist: change almost nothing. Regular users know exactly where everything is. Stability is a feature for a tool used for years. Cost: missed improvements (mobile ergonomics, safety features) that rivals exploited.
8. Craigslist: business model shapes design. Most listings are free; fees apply to a few categories in some cities. Without an ad business, there is no reason to add engagement features. Cost: a smaller company that innovates slowly.

Trade-offs and criticisms

Google's homepage is still minimal, but the results page has become one of the most contested designs on the web: ads that look like results, then panels, then AI summaries that answer without sending traffic to the sites they draw on. Minimal entry and maximal results are now in tension. Craigslist's refusal to change left openings for competitors: Facebook Marketplace, OLX and Quikr in India, and many category-specific apps took share with photos, profiles and safety features. Appropriateness is not permanent. The job changes, the devices change, and so must the form.

Related products

• Hacker NewsY Combinator, 2007
  A dense, text-only list that barely changes. Craigslist's lesson in a community product.
• DuckDuckGo2008
  A search engine that kept results simpler and made privacy its edge. Fit can be a position against the leader.
• OLX, Facebook Marketplace2006; 2016
  Photo-first classifieds with profiles and chat. A changed job (mobile, visual goods, trust) needed a changed form.

Learning a language takes hundreds of hours. Most people quit within weeks. Duolingo's design bet is that the hard part is not teaching, it is coming back. So the product makes coming back easy (lessons of a few minutes), rewarding (XP, sound, animation) and costly to skip (a streak you do not want to lose, an owl who reminds you). The company tests constantly and has written publicly about how the streak became one of its most important features. It also uses spaced-repetition research to schedule review of words you are likely to forget.

1. The streak. A count of consecutive days with at least one lesson. It turns a vague goal (learn Spanish) into a concrete daily one (keep the number going), and loss aversion does the rest. Cost: people do the minimum to protect the number, and a broken long streak can make someone quit entirely.
2. Streak protection. Streak freezes (and similar features) let a day be missed without loss. They reduce the cliff of a broken streak. Cost: they also make the streak something you can buy, which blurs the line between motivation and monetization.
3. Lessons of a few minutes. The unit of commitment is tiny, so starting is easy. Cost: depth; real conversation needs longer, harder practice.
4. One path, one next step. The home screen is a single path; the next lesson is obvious. No choice paralysis. Cost: less learner control.
5. A mascot with a voice. Duo the owl delivers reminders, jokes and guilt. Personality makes nagging tolerable, even shareable; the "menacing owl" became an internet meme that the company embraced in marketing. Cost: the joke depends on a relationship built on mild pressure.
6. Leagues and social comparison. Weekly leaderboards with promotion and demotion. Cost: extrinsic motivation can crowd out intrinsic interest; some users grind easy lessons for points.
7. Hearts and freemium limits. Mistakes cost hearts in the free tier; the subscription removes limits and ads. Cost: the learning mechanic (mistakes) is tied to a monetization mechanic (limits), which can punish exactly the behaviour learning needs.
8. Testing everything. Many changes ship as experiments measured on retention. Cost: what is measured (daily activity) can drift from what matters (actual language ability).

The ethical debate

Duolingo's mechanics clearly increase daily practice, and daily practice is good for learning. Critics argue that the mechanics optimize for the streak and the app, not for fluency: people complete easy lessons to keep a number alive, and guilt-based reminders train anxiety. Defenders reply that without the mechanics, almost nobody would practise at all. Both are partly right. The useful question for you is not "is gamification bad?" but "does each mechanic reward the behaviour that actually achieves the user's goal?" A streak for a daily habit the user chose is a tool. A streak that keeps someone tapping after the habit stopped serving them is a trap. Read Honest & Calm alongside this chapter.

Related products

• AnkiDamien Elmes, 2006
  Spaced repetition with almost no gamification. Very effective for motivated learners; very hard for everyone else.
• Headspace, Calm2010s
  Meditation apps with streaks and gentle reminders. The same mechanics in a domain where guilt is especially counterproductive.
• Snapchat SnapstreaksSnap, 2015
  Streaks between friends. A widely criticized case of streaks creating social obligation.

WhatsApp became the default way India communicates, not because it was beautiful, but because it worked: small, fast on cheap Android phones, tolerant of bad networks, identity by phone number (no username to invent), and voice notes for people more comfortable speaking than typing. UPI did the same for money. You open any UPI app, scan a QR code at a tea stall, type ₹20, enter your UPI PIN and see a big tick. The shopkeeper's speaker box announces the amount out loud. In a few years, this became how India pays, with billions of transactions a month.

1. Ticks make the network visible. One grey tick: sent to WhatsApp's server. Two grey ticks: delivered to the phone. Two blue ticks (added in 2014): read. In places where networks drop and phones die, knowing whether a message arrived is half the value. Cost: read receipts create social pressure, so they can be turned off (which also hides others' receipts from you).
2. Phone number as identity. No username, no password at sign-up; your contacts are already there. Cost: privacy, since your number is exposed to every group you join.
3. Voice notes as first-class. Hold to record, slide to cancel, lock to record hands-free. Voice works across literacy levels, scripts and languages. Cost: long voice notes are hard to skim (WhatsApp later added faster playback and transcripts).
4. Built for cheap phones and bad networks. Small app, efficient data use, messages queue and send when the network returns. Cost: engineering constraints on features.
5. Friction against misinformation. After viral rumours spread through forwarded messages, especially in India, WhatsApp labelled forwarded messages (2018) and limited how many chats a message can be forwarded to at once. Deliberate friction as a safety feature. Cost: inconvenience for legitimate sharing.
6. UPI: one decision per screen. Scan, confirm amount, enter PIN, see result. Each screen asks one question with one large button. Cost: little room for extras, which is correct.
7. UPI: a standard PIN step. PIN entry is a standardized, bank-backed step that looks much the same across UPI apps, which builds familiarity and trust whichever app you use. Cost: apps cannot customize it; that is the point.
8. UPI: success that is unmissable, for both sides. A full-screen tick and sound for the payer, and a voice announcement on the merchant's speaker box, so a shopkeeper does not need to look at a phone. Cost: hardware and audio in a noisy shop; and fake "success" screens became a known scam, which is why the merchant's independent confirmation matters.

Trade-offs and criticisms

Read receipts gave people a new source of anxiety ("they read it and didn't reply"). Phone numbers as identity expose users in large groups. WhatsApp's very success in India made it a channel for misinformation, and its fixes are friction, not cures. Meta's ownership raises privacy concerns, even with end-to-end encryption of message content (since 2016). On UPI, the ease of paying invites scams: requests to "collect" money that people approve by mistake, and fake success screens. Every simplification shifts the burden somewhere; good design moves it to where it can be handled, like an independent audio confirmation for the merchant.

Related products

• SignalSignal Foundation
  Similar ticks and simplicity with a stronger privacy stance and usernames. The same interaction model with different values.
• M-PesaSafaricom, Kenya, 2007
  Mobile money over basic phones using SMS menus. Inclusion through the most basic interface available.
• PhonePe, Paytm, BHIMIndia, 2016–
  UPI apps with the same core flow and different extras. When the rails are shared, the core flow converges; differentiation moves to the edges.

Seen together, the products show that great software is coherent rather than perfect: each chooses one bet, makes every layer serve it, and pays openly for what it refuses.

Read the table across a row to see one product's logic: its bet, the decision that expresses it, the business model that funds it, the thing it refuses and the cost it accepts. Read down a column to compare. You will notice that the refusals column is as telling as the signature column. Then use the pattern index when you face a design problem: find the principle, then study the products that exemplify it.

Pattern index

When you face a problem in your own product, start here. Each row names a principle, the breakdowns that show it best, and the guide that teaches the underlying theory.

In the wild

• Product clonesEvery year
  Many launches each year are "X for Y" clones that copy a famous product's visuals into a different job. The ones that last translate the bet, not the look.
• Things vs OmniFocusLong-running rivals
  Two coherent bets (calm versus power) for the same category, both sustained for over a decade. Coherence beats feature counts.
• Gmail adopting split inbox ideas2010s
  Incumbents absorb good patterns from bold products. A pattern alone is not a moat; a coherent whole is harder to copy.

The point of reading breakdowns is to become someone who writes them: a teardown a week, done with the protocol from chapter 2 and this template, will train your eye faster than any amount of browsing.

Every chapter in this guide followed the same skeleton: facts, the core bet, a pinned screen, a numbered list of decisions with costs, trade-offs and criticisms, things to steal, and a test of understanding. Use the same skeleton. Fill it with your own product choices, and keep your teardowns in one place. After ten of them, you will see patterns that no article could hand you, because they come from your own eye and your own questions.

The template

Copy this into your notes app (it is plain Markdown, so it works in iA Writer, Bear, Obsidian or a text file). Fill one per task.

# Teardown: [Product] · [Task in the user's words]
Date: [YYYY-MM-DD] · Platform: [iOS / Android / Web / Mac] · Starting state: [fresh install / returning user with data]

## A. The bet (one sentence)
[This product believes that ...]

## B. Facts
- Maker / team size:
- Year and version:
- Who pays, and how:
- Main rival I compared against:

## C. Screens
[Strip of numbered screenshots: 01-..., 02-...]

## D. Decisions (one per pin)
| # | Decision (what) | Reason (why I think) | Cost (what it gives up) | Confidence |
|---|-----------------|----------------------|-------------------------|------------|
| 1 |                 |                      |                         | high / med / guess |
| 2 |                 |                      |                         |            |

## E. The nine lenses (one line each)
1. Purpose / job:
2. Conceptual model:
3. Structure:
4. Interaction:
5. Visual and type:
6. Feel and motion:
7. Words:
8. Business model's effect:
9. What is left out:

## F. Trade-offs and rival
- Who this design fails:
- Three differences from [rival]:

## G. Steal
- Patterns worth borrowing (and does each fit MY bet?):
- My one change this week (specific, shippable):
- How I will know it worked:

A worked micro-example

Here is section D from a short teardown of one task: "add a rupee expense while standing in a shop", in a hypothetical expense app compared with typing it into Apple Notes.

Notice the confidence column. Being honest about guesses keeps a teardown from becoming fan fiction about a product's intentions.

In the wild

• Designers' public teardownsBlogs and newsletters
  Many designers built audiences with regular, careful breakdowns of apps they admire. Consistency matters more than length.
• Lux's camera reviewsHalide blog
  Each iPhone generation gets a deep, honest teardown of its camera. Teardowns can be your product's best marketing.
• Apple Design Awards and Behance case studiesAnnual
  Winners' and authors' own write-ups explain their decisions. Compare their stated reasons with your teardown's guesses.

Wrap-up

Practice assignment for the next two weeks

1. Days 1–2: Set up. Create your teardowns/ folder and paste the template. Write your own product's row for the comparison table (exercise 21.1).
2. Days 3–4: Admired product. One-task teardown of a calm app you love (iA Writer, Things, Bear or similar). End with one change.
3. Days 5–6: Frustrating product. One-task teardown of an app that annoys you. Find at least two decisions that are actually reasonable from its business model's point of view.
4. Days 7–8: Direct competitor. Tear down the same task in your closest competitor and in your own product, side by side.
5. Days 9–10: Ship. Ship the best one of the three changes you collected. Measure something simple before and after (time on task, taps, support emails).
6. Days 11–12: India test. Run your core flow on a low-end Android phone on a throttled network (exercise 20.1), and tear down your own app's result.
7. Days 13–14: Publish and reflect. Polish one teardown and publish it with credit. Write a half-page reflection: which lens do you now see easily, and which is still your blind spot from exercise 1.1?

Reading list

• The Design of Everyday Things (revised edition, 2013)
  Don Norman · the vocabulary of conceptual models, feedback and mapping behind every lens in this guide.
• The Elements of User Experience (2002; 2nd ed. 2010)
  Jesse James Garrett · the five planes, a clear model of how abstract decisions shape visible ones.
• "Web Design is 95% Typography" (2006) and the iA blog
  Information Architects · the thinking behind iA Writer, including essays on focus, fonts and AI authorship.
• "How Superhuman Built an Engine to Find Product/Market Fit" (First Round Review, 2018)
  Rahul Vohra · the 40% survey method in detail, with the segmenting and roadmap logic.
• The Linear Method
  Linear · a short, opinionated set of practices that the product encodes; read it next to the app.
• "File over app" (2023)
  Steph Ango · a short essay on durability and ownership, the philosophy behind Obsidian.
• Stripe API reference: Errors
  Stripe · a model of error design you can copy for any product with integrations.
• "How Do Users Really Hold Mobile Devices?" (UXmatters, 2013)
  Steven Hoober · the research behind thumb zones and Halide-style reachability.
• The Halide blog (Lux)
  Sebastiaan de With and Ben Sandofsky · deep camera teardowns; a model of teaching your craft in public.
• marco.org
  Marco Arment · years of posts about building and pricing Overcast alone; essential solo-creator reading.
• Hooked (2014)
  Nir Eyal · the habit loop explained by an advocate; read critically, alongside the next item.
• Calm Technology (2015)
  Amber Case · principles for technology that informs without demanding attention; the counterweight to Duolingo-style loops.