Introducing Fin: The world’s first AI Chief Financial Officer. Fin outperforms humans 100% of the time. RT + Comment “FIN” and I’ll send you an AI agent that saves 6-7 figures/year.

"Portrait orientation" has historically meant "a tall image", in contrast to "landscape orientation". In 2016, Apple released "portrait mode" for iPhone, which blurs out the background of an image (I developed most of Google's equivalent tech from 2013-2015, we called it "lens blur" initially and then adopted "portrait mode" in 2017). Since then, vertical video has become ubiquitous and background defocus has become an invisible part of many mobile photography pipelines, and as such "portrait mode" has slowly absorbed the meaning of "portrait orientation" to the point where it has now almost completely displaced it! Watching this slow linguistic shift over the past decade has been a strange experience for me.

We also built a spreadsheet agent similar to @tryshortcutai. Spreadsheet UI on the left, AI chat on the right. The agent has database tools - query, read, write, delete - and code execution built in. Ask it to create tables, run calculations, or add formulas. It executes the operations and updates the sheet directly. The agent decides what database operations or code to run based on your request.

Most companies are *vastly* overcomplicating their analytics. Everything is tracked clicks, scrolls, impressions, events. Which is fine. Logging is cheap. We also need them when we need to understand rare phenomena. But attention isn’t cheap. Most of what we track never helps us make better decisions. The truth is, only about 100 metrics really matter. These 100 metrics explain 90% of what’s happening in your business and product. And the same principle holds elsewhere too: Only 50 events truly matter for understanding user and system behavior. Only 150 entity characteristics — the key attributes of your users, products, or content — explain most outcomes. Everything else lives in the long tail: useful for special cases, but not essential for running the business on a daily basis. This is because everybusiness can be represented as a system, and these systems can be written as a set of equations. When you express your business as equations, you expose its levers. These levers are potentially actionable and can actually move results. Take Facebook’s revenue model. It can be simplified into four components: 1. Revenue = Users × Impressions per User × Ad Impressions per Impression × Revenue per Ad That’s it. Four levers at the highest level. To grow revenue, you can: 1. Increase users (growth) 2. Increase engagement (more impressions per user) 3. Raise ad load (more ads per impression) 4. Improve monetization (revenue per ad) Each of these can be broken down further. Let’s choose Monthly Active Users (MAU) as a proxy for growth. You can decompose MAU by an equation. MAU = New Users + Retained Users + Resurrected Users You can also grow your active users by getting new users, resurrecting churned users and keeping the existing users from churning. Now, let’s go one layer deeper. New Users = Visitors × (Downloads / Visitors) × (App Opens / Downloads) × (Registrations / App Opens) × (New Users / Registrations) If we define a new user as someone who registers and then takes an action, growth comes from improving each step of that journey. We can bring in more visitors at the top of the funnel, get more of them to download the app, increase the share who open it, raise the percentage who register, and finally help more of them take their first action. Each step is measurable. Each can be improved. Each has a story behind it. And if you want, you can keep peeling — looking at funnel drop-offs, activation, or engagement drivers. This is the beauty of decomposition. When you break the system into equations, you can see what drives what. After you do this for the key levers of your business, add all your metrics up. I'd be surprised if what *truly* matters is more than 100 metrics. More on our latest post in Opinionated Intelligence dot substak.

I always do a SELECT * to double check the entries before switching it to DELETE (usually with a limit)

Still thinking about this chart, wherein NBA is most concerned w/optimizing shots - at cost of entertaining gameplay - and since doing this viewership is down significantly. Is what happens when you let spreadsheet-brained NPCs run everything -music, movies etc all made boring

Four decades in, some things never change … Then and now.

Most product teams live and die by analytics. We obsess over every click, every drop-off, every chart on the dashboard. But ask anyone what their users were actually thinking when they left and you’ll get silence. Every company has the same blind spot. Data tells you what happened. It never tells you why. Now imagine this: the answer was hiding in a forgotten tool. Nearly every founder once used it. Then it disappeared. Back in 2009, analytics set the pace for every product decision. Heatmaps tracked every tap and funnel, but no tool could explain what users felt in the moments that mattered. Everyone in product knew the gap. Interviews took too long and reached too few. Traditional surveys were exhausting and went unanswered. That gap became our mission. Alongside Sean Ellis, we built a survey tool to assess product/market fit. That tool was just the starting point. We built a way to embed the whole product/market fit survey on a website. That was the original inspiration that started it all. The tool evolved into an entirely new product. What changed everything was giving teams a way to listen directly, right as decisions happened. The survey appeared on the website, not in an inbox. Thousands of startups started using the tool to catch answers in real time. A SaaS founder could ask, “What almost stopped you from signing up?” and see insights appear before the next deploy. E-commerce brands dropped, “What made you leave your cart?” right at the point of hesitation and saw patterns emerge in a single afternoon. One team learned their pricing page was confusing after hearing it from dozens of users in just a week. Another caught the phrase “missing features” so many times that their roadmap shifted overnight. The lesson was simple. Context is everything. If you ask at the right moment, even the quietest users will tell you what matters most. We wanted to go even further. Early prototypes were scrappy. Hand-drawn boxes pasted on screenshots, phone calls with a single question, landing pages that promised, “Ask your users in 30 seconds.” Every version focused on one goal: make it effortless to ask a question right at the point of friction. KISSinsights became the purest form of this idea. Teams dropped a small script onto their site. Instantly, a question box appeared wherever it was needed. There were no settings to configure and no training required. Founders put surveys on pricing pages to ask, “What’s missing before you pick a plan?” Marketers added them to sign-up flows, asking, “What almost stopped you from joining?” Product teams set a survey for users about to leave, asking, “What didn’t you find today?” The answers landed instantly in the team’s inbox, often within seconds of the user’s decision. What surprised us was how users responded. They answered, and many even thanked the team for caring enough to ask. One customer wrote, “This is the only pop-up I actually want to see.” Another founder said they learned more in two days from KISSinsights than from months of analytics reports. By meeting users where they were, teams finally heard honest feedback from people who would have otherwise stayed silent. There was an unexpected growth loop, too. Every survey ended with a simple line: “Powered by KISSinsights.” Curious users clicked, set up their own micro-surveys, and the idea spread. A founder who saw the pop-up during checkout added it to their own launch page a week later. Teams swapped stories in forums and group chats about the best questions to ask. One growth marketer shared how a single survey on the pricing page led to a thirty percent jump in conversions after they fixed a confusing headline. Within months, more than twenty-five thousand teams were using KISSinsights. It spread founder to founder, blog to demo, and conference to Slack channel. For a time, micro-surveys became a default part of every serious product and marketing team’s workflow. The best teams made it a habit to ask a real question at the moment of truth, every week. But change came quickly. Voice-of-customer suites and enterprise vendors took over. Free plans disappeared or became heavily limited. The simple experience was replaced with dashboards and paywalls. Multi-step surveys became the norm. The original discipline of ask one real question, at the exact right moment, was left behind. What was lost went beyond speed or simplicity. Teams lost the habit of meeting users in context. Today, most teams have more dashboards and less clarity. The most honest answers that surface the why in the user’s own words have almost vanished. If you want to know why people leave, or what’s stopping them from buying, you still need to ask the right question in the right moment. Micro-surveys were about making it so easy to listen that no team could ignore the answers. It’s easy to forget how much a single, well-timed question can teach you. But if you built products in those years, you remember the little box that changed everything.

Josh Allen & Tua Tagovailoa combined to average 1.5 air yards per completion, the lowest mark in a game in the NGS era (since 2016). Allen attempted only one pass over 10 air yards, while Tua completed as many passes behind the LOS as past it (11 each). Powered by @awscloud

As a neurosurgeon I care a lot about road safety. By now you’ve probably seen @Waymo’s stunning safety results (like 91% fewer serious crashes). But they didn’t just publish data headlines. They released the raw CSV files and data dictionaries. I did a much deeper analysis. A fascinating story emerges when you analyze how they’re achieving this. This isn’t incremental improvement - it’s categorical. We’re looking at the potential elimination of traffic deaths as a leading cause of mortality. The intersection breakthrough: Waymo has essentially solved intersection crashes, with 95% fewer injury incidents than human drivers in the same locations. That’s transforming the deadliest driving scenario. The national math: If every US vehicle performed like Waymo, we’d prevent 33,000-39,000 deaths annually and save $0.9-1.25 trillion in societal costs. Even partial adoption at 27% would save ~10,000 lives per year. In terms of magnitude, this would be the equivalent of eliminating every pedestrian death nationally in a year. The physics signature: Here’s what fascinates me: 47% of Waymo’s contacts involve less than 1 mph delta-V. They’re not just avoiding crashes; they’re converting unavoidable incidents into gentle bumps. It’s like having physics itself on your side. We’re not talking about marginal safety gains. The data represents a fundamental shift from harm reduction to harm prevention. The methodology matters: I used their dynamic geographic benchmarks (comparing like-for-like road conditions) and verified the findings hold across San Francisco, Phoenix, LA, and Austin. The safety advantage actually increases in more complex urban environments. Link to raw data below…. Notes on my approach: Analysis based on 96 million miles of Waymo Rider-Only (RO) data through June 2025, utilizing Waymo's dynamic geographic benchmarks to compare Waymo Driver performance against human drivers under similar road conditions and operational design domains. The projections for national impact (deaths prevented, societal costs) involve several assumptions. Given Waymo's zero reported fatalities, the direct serious injury reductions were mapped to national fatality statistics using established NHTSA-derived ratios that correlate serious injury crash rates with fatality rates. This extrapolation assumes that Waymo's observed serious injury prevention capability would translate proportionally to fatality prevention. Societal cost savings are estimated by applying average per-fatality and per-injury economic costs (e.g., medical, lost productivity, quality of life) as published by NHTSA, scaling these national averages to the projected number of avoided fatalities and injuries based on Waymo's safety performance. These figures represent the potential annual impact if the Waymo Driver's safety profile were widely integrated into the national fleet. @ethanteicher

This is the entire google SEO algorithm in 1 screenshot:

Hear me out... GitHub activity graphs but for tweeting (reply and I'll send a screenshot of yours)

There is no word in English that means someone is 30% sure something will happen

WHY RETENTION IS SO HARD FOR NEW TECH PRODUCTS I’ve been staring at retention curve data for 15-plus years now. I was a founder, a product manager, and now a VC. And at Andreessen Horowitz, I end up meeting hundreds of startups each year, many of them through our a16z speedrun program, where we invest up to $1M in brand new startups. NOTE: And yes, we just announced the 2026 program and you can apply now: speedrun006.a16z.com/x But back to retention — I’ve seen thousands of curves, and it’s among the first things I ask for when evaluating a new startups. I've looked through thousands of data rooms, analyzed retention curves sliced across many segments and denominators. I've also seen it from the other side, as someone building products. I’ve run hundreds of A/B tests and drafted countless variations of onboarding and notification emails in attempts to bend the curve. There are patterns. Just as there’s the laws of physics, weirdly there are some constant patterns that keep cropping up over time. Here are a few that I’ll share: - You can’t fix bad retention. No, adding more notifications will not fix your retention curve. You can’t A/B test your way to good retention - Retention goes down, it doesn’t go up. And weirdly, it decays (oh, does it decay) at a predictable half life. Early retention predicts later retention. - Revenue retention expands, while usage retention shrinks. Good news: You lose people over over time, but the ones that remain sometimes spend more more money! - Retention is relative to your product category. There’s nature, and there’s nurture. Sorry, you’ll never make a hotel booking app a daily use product - Retention gets worse as users expand and grow. The best users are early and organic. The worst users come after that - Churn is asymmetric. It’s far easier to lose a user forever than to re-win them back - Retention is weirdly hard to measure. Seasonality is a real thing. New tests throw things off. Bugs happen. D365 is a real metric but you can’t wait - Crazy viral growth with shitty retention fails. We’ve run this experiment many many times already, across multiple platforms and categories - Great retention is magic. When you see it out in the wild, it’s amazing. We’ll dive into each of these. You can’t fix bad retention. You've seen this happen before: You spend months developing a new product, and then you launch it. Bad news hits. The initial retention stats come in, and it’s terrible. You're already months into the product development, and it's hard to turn back now. How do we improve retention? I know, let’s add notifications and remind people to come back. Let's add a bunch of new features. Maybe we can A/B test the landing page and increase conversion. I think we know how this ends. Unfortunately, it seems to be the case that when you have poorer retention, it's very, very hard to fix - nearly impossible. Yes, you might be able to make a marginal improvement. Let's say that your D1 retention is 40% and you'd like it to be 50%. This is great and potentially workable. If the D1 is 10% on the other hand, well, you probably just have built something that nobody wants, and all the local optimizations around A/B tests and notifications aren't actually enough to bend the curve enough for it to work. When there's been months of development and sunk cost, it's hard to not just give it a college try. But I think in many cases new products are better off pivoting right away. The type of pivot that fixes retention involves a complete new redesign of the app's home screen. If it looks like a feed, maybe it needs to be a structured step-by-step flow. If the product is about sharing, maybe it needs to be mostly about creating and saving. You might need to describe the product in a totally different way and position it against a different product. It needs to be a big pivot in many ways, the bigger the better, in order to have a chance at changing retention. Retention goes down, it doesn’t go up. Retention curves often follow very geometric curves that you see. For instance many curves I see resemble the following: Whatever the D1, it drops by 50% on D7. Whatever the D7, it drops by another 50% at D30. Months out you might end up at roughly zero, or if you’re lucky you might retain 10% overall. There’s just a predictable decay. What you never see is a curve that starts high, then goes low, then becomes high again. That’s not possible. In other words, if your early retention isn't incredibly good, then it means that your late retention also probably isn't any good. You need to start strong in order to end strong. There are some interesting exceptions to this rule that are worth calling out: - Some products are extremely hardcore (e.g. online poker). You might have relatively low retention, but those who stay are extremely sticky and spend a ton of money. It turns out that this can work. - A product that has network effects where new users might start out strong, then drift for a bit. But if the product (which might be a social network or a collaboration tool or something else that has network effects) is able to use more and more users to reactivate older users, you often see a small curve where retention comes up. This is a very rare type of situation but amazing when possible. Revenue retention expands, while usage retention shrinks. One of the best and most important dynamics with retention curves is that you can apply them to users, but then also revenue. Thus far, we've been talking about user retention, and unfortunately, it has the undesirable dynamic of always going down. Revenue retention on the other hand is really interesting because people often end up spending more money over time with you, at least the ones that remain. This is one of the biggest strengths of B2B SaaS products. Take a product like Slack. If you look at the user cohorts, what you'd likely find is that the retention curves go down just like any other product. Some people take to it, and some people don't. However, for the companies where people spend time adopting Slack, what happens is it will start to organically grow, and the amount of revenue you earn from that company starts to increase dramatically over time. The revenue retention curves start to grow rather than shrink. This is amazing, and unfortunately doesn’t apply to most consumer products. It’s one of the biggest ways in which B2B products have easier business model dynamics than consumer. The consumer version of this looks more like Amazon where you might have started by buying books and music and over time as the product grows in its capabilities you start to use it to buy more and more things. Because of that your LTV in the product is essentially unbounded. We also saw this at Uber as well where user cohorts would decay over time but the amount of money that people would spend initially on Uber rides to the airport would grow into rides to restaurants or for commuting purposes to work. So the user retention curves go down but the revenue retention curves go up. Retention is relative to your product category. I've written about this in the past on the concept of nature vs. nurture for retention. The reality is that there's just a natural use case for many products - for example with collaboration tools or coding apps, you might use them every day at work, capping your usage to 5 active days out of 7. Contrast that to something like a bug alert system - hopefully you don’t use it often! Same with consumer products, where people check news, messaging, and social apps daily, but generally don’t use medical reference guides frequently. Some apps have great retention but infrequent usage, like weather or banking apps. And some categories like gaming are highly addictive and frequent, but people usually quit after a few weeks of use once the content is played out. Nature vs. nurture is important because it tells you that many new products simply don't have a chance. If you're developing a travel app, but it's meant to be social, the reality is people don't travel that much. It'll be hard to create a product that sole mission is interaction with friends. Instead, it would be better to accept its infrequent nature and figure out how to monetize it better by owning part of the transaction or to have a more frequent use case like a restaurant and nightlife app like Yelp but also be able to use travel features as well. It's just hard to fight nature. You can only do so much. It's also for this reason that if you want to build a very, very high retention, high frequency app, you have to probably build within some of the categories that people are already identifying as core daily products. This means that most likely if your app is successful, it takes away from some other daily product. It's no wonder that my constant use of ChatGPT has dramatically diminished the number of Google searches that I do. Or that when I began to use Substack for reading and writing blogs that I stopped using many other kinds of social news software. Retention gets worse as users expand and grow. If you are lucky enough to build a product that experiences great retention, one of the natural things is simply to extrapolate all the behavior, monetization, and usage to a much broader market and assume that naturally you end up with a very, very big good number because you're multiplying a bunch of small good numbers together with a big one. The reality is that as you start to scale your user base, bad things start to happen. Let’s say you begin adding Android and international users and you acquire more customers using paid marketing and other channels — you'll quickly find that all of your metrics get worse. The reason is that the best users show up early. The ones that are the most highly monetizable, that have the highest intent, that are the most digital and plugged in, well, these guys tend to find your product early and start using it due to a recommendation from a friend. Later on, as you bring in new users from other sources, it's likely that your product just isn't as good for them. It could be as simple as building an iPhone app for college students in Western countries and simply getting worse metrics as you bring in Android users from emerging countries where the feature sets just don't quite work. Of course you can work to improve this over time, but I assure you it will never be the same. Instead, the question is: As you grow your users and they get worse and worse, are they still valuable and can you still operate your product profitably? And more importantly, are you able to hold on to that core highly valuable user base that came in early? No wonder these early users are often called The Golden Cohort. Churn is asymmetric. It's incredibly easy to churn users. In fact, most products churn 90% or more in the first 30 days. Simultaneously, it's incredibly hard to win back a user that's already quit. This is the core asymmetry around churn. In fact, it's so bad that it's often easier to simply try to acquire a new user rather than to try to get someone back. It's for this reason that life cycle marketing that involves trying to resurrect dormant users by sending them discounts or offers tends to be extremely painful and expensive. The version of this that often works is to get existing engaged users to resurrect somebody through the natural usage of the product. For example, if somebody at work tries a new project management tool and it doesn't stick, then you probably won't get them back by bombarding their emails with reminders of features. Instead, you try to get one of their coworkers to invite them back into the tool to work on a new project. That's what works. But again, insanely difficult and complex and is really only available to products that have network effects (i.e., sharing and collaboration). Retention is weirdly hard to measure. When people talk about retention, they tend to try to measure what happens in the first day, first week, and first month. But they'll rarely talk about what happens two years ahead. The reason is that when you're working on a product, you need a short enough time frame and a thing that's easy enough to measure that teams can make decisions about what is happening. As a result, although annual churn or long-term monetization is incredibly important, you tend not to measure it, instead focusing on what's right in front of you and what's easy. However, this approach has many problems. Unfortunately, many categories of products experience huge amounts of seasonality. Anything involving commerce, travel, wellness, or online dating are obvious examples. But there are cycles even to the way that companies use business software as well. Seasonality throws things off because you might be down month over month or quarter over quarter, but is that because of features that you launched? Or is it because user behavior is simply different in this quarter? It's just hard to measure retention when it's super laggy. Same with bugs or new tests that you're running or new market launches. These are all things that muck up the data, and you end up finding yourself reviewing reports where retention curves went up or down. But there's an asterisk on every number because they're trying to validate that the new Android launch didn't create an apples-to-oranges comparison. Crazy viral growth with shitty retention fails. Many folks working on new products find themselves very focused on signing up new users and not on retention at all. After all, if you just want to see a graph that goes up and to the right, why not simply ramp your top of funnel and show that you're growing quickly, raise a bunch of venture capital money, and then you can figure out the retention issue later. We're seeing this all the time right now in the industry when products have a crazy TikTok ramp because a creator pushed their app to millions of followers or because a launch video caused a bunch of revenue growth. Even though the usage and churn are not in a good place. The tech industry has already run this experiment many, many times. And the conclusion is the same: Highly viral products with shitty retention do not last because it's so hard to fix retention. Eventually, the user acquisition fades as the novelty factor goes away, and eventually you're left with shitty user acquisition and shitty retention, and what goes up must come down. We've seen this across many contexts. During the early social network phase, there were many products that used email address books to spam their way to growth, but drove users to bad products. Sometimes, if you could get them to sign up to some shitty ringtone annual subscription, you could try to monetize them and make some money along the way. It wasn't until Facebook, of course, with their UX innovations like The Feed and Real Names, that eventually created a product that was both highly viral and had very high retention. The same thing has happened in mobile apps as well, where you see big hits pop up sometimes caused by forced invitations via SMS, but again, if the products aren't sticky, the whole thing collapses quickly. Great retention is magic. You might read this whole essay and feel a little bit depressed. I know that sometimes it's hard to get things going. However, it's amazing when something really works. When you see a product out in the wild with a 50% D30 (I do once every couple years), it's just amazing. I've come to believe that these lightning-in-a-bottle products happen not because the builders had some incredibly systematic way to A/B test their way to great metrics or that they employed some kind of high-velocity iteration process that got them there, but simply that there's a little bit of magic that's required. This magic comes from a fresh insight about the market or customer needs, and while it might seem obvious in retrospect, it drives super high retention because this product is the first to figure it out. We can say this now about video conferencing software or disappearing photos or a magic AI that replies back to you about any topic. There's just a magic here that no amount of iteration and metrics-driven testing can get you to. The Big Question You might read all of this and still have a big question: So wait, how do you get to great retention? (If I knew the answer in a deterministic way, my job as a startup investor would be so much easier, wouldn’t it?) But let’s try our best. In my points above, there’s a few clues: The idea really matters. - If you want a high retention product, you need to pick a category that is high retention already. - You need to pick a product category where you already use an existing product every day. - You're going to build something that directly competes against that. - If you win, then you'll stop using that other product and use your product instead. - That's a high bar, but I think it's a good start. Of course, if you build something that's quite head-to-head with something that already exists, you might suitably object: "that's going to be really hard to switch somebody over." It is. So then this is where you need to decide to take enough market risk, but just the appropriate amount, where you do something new and different that reinvents that core interaction. But you're probably talking more about a 20% remix rather than an 80%. Ideally, you need to be able to describe this to your users in a way that they can understand quickly and viscerally within the first 60 seconds of usage. This is where the dreaded investor question "why now?" starts to matter quite a bit. Because what you're saying here is that ideally there's some kind of new development in the industry, whether that's a general-purpose technology like LLMs or a societal difference like the oversaturation of social media, that allows you to make this twist happen at exactly the right moment. This gets you into an existing market quickly, and you're more likely to have great retention numbers early on. Timing matters a lot. If you get the timing off, and it's a low-interest category, and your differentiation isn't different enough, then what you'll find is you've traded a retention problem for a user acquisition problem. Here's the difficulty with building a new kind of internet browser: if you win, it's incredibly sticky. But people are so happy with their existing browsers that it's very expensive and complex to get them to try yours in the first place. This is why I don’t blame folks who have a “Cursor for X” idea, or “Figma for X,” just like the “Uber for X” ideas of the past generation. They’re trying to piggy off some existing markets and behavior so that they don’t have to take crazy market risk. And if you get the differentiation right, the timing right, and there’s a ton of user demand, and you’ve nailed the right base product category, then I think it can really work. But what about new markets? The natural counterpoint is that new markets are often more exciting than existing ones. Isn't tech about building brand new things rather than innovating 20% on old stuff? Of course this is true, but I think this is the tiny tiny minority of products. My counterpoint to this counterpoint is that most products actually have some kind of prior lineage, even if those prior products are quickly forgotten. Before Instagram there was Hipstamatic, which had become the #1 paid photo app in the early App Store. It demonstrated the success of photo filters. Of course Google was not the first search engine, it was actually #10 or whatever, after Lycos, Excite, Infoseek, etc., which demonstrated consumers wanted search but that it was impossible to monetize. Tesla was not the first electric car, nor iPhone the first smartphone. Sometimes it’s the 10th iteration that matters. Some call this “last mover advantage” rather than first mover. I think an important point. Yet sometimes new things do happen. Uber was created to turn an existing offline action — calling a cab — into an app, not because there was already a hugely successful ridehailing app. (And no, not Lyft — it was a weird bus booking thing at the time). Of course a lot of ChatGPT, with OpenAI’s 5 year journey between inception and v3 which really took off, and without any real blueprints for what it might replace. These types of journeys are remarkable, and the tech industry is better off for it, because they involve real risk as part of new category creation.