Search the archive

LAYOUT DESIGN SYSTEM FOR THE AI ERA

Kami is a layout design system for the AI era that makes documents clear, readable, and memorable.

PROCEDURAL UI SOUNDS FOR THE WEB

tiks generates UI sounds (click, toggle, success, and error, etc.) at runtime using oscillators and gain envelopes through the Web Audio API. ~2KB gzipped.

DESIGNING FOR PEOPLE WITH READING DISABILITIES

Reading disabilities like dyslexia, hyperlexia, and alexia affect how the brain processes written text — not intelligence — making design choices crucial for reducing cognitive load. Plain language, left-aligned short lines, distinct typography, and descriptive link text are among the key recommendations for digital content that works for more people. These principles benefit not just those with reading disabilities, but also second-language readers, people in a hurry, and those with lower digital literacy.

01-20 DESIGNS THE WEBSTER'S BRAND ON PRESENCE, WONDER, AND CONNECTION

The Webster's new brand identity, created by 01-20, is built around the idea of "the magic of presence," using the subtle wonder of magical realism—particularly inspired by Gabriel García Márquez—to make hospitality feel more human, emotionally engaging, and culturally rooted rather than transactional or overtly luxurious. The identity combines literary-inspired typography, warm earthy colors, carefully used illustrations, photography, and local storytelling to create meaningful moments while staying grounded in each hotel's architecture, history, and community. Rather than enforcing a rigid visual formula, the brand acts as a flexible storytelling framework, encouraging each property to express its own character through cultural programming, design, and connections to its neighbourhood.

INTRODUCING OPENART DIRECTOR

OpenArt has introduced OpenArt Director, a tool built around "Vibe Directing" — a concept positioned to change video creation the way vibe coding changed software development. Users can describe their vision conversationally and receive videos up to 5 minutes long, complete with consistent characters, branding, voiceover, music, and captions — all without editing timelines. The platform supports 8+ languages natively and handles everything through a single chat interface, from initial creation to edits.

SUPERMAN POSTERS: THE HEROIC LEGACY IN STUNNING COMIC ART

Superman's legacy as one of the world's most recognizable superheroes spans over eight decades, dating back to his comic book debut in 1938.

BEHOLD, THE COOLEST MODERN APPLE IPOD CONCEPT YET

A modern Apple iPod concept featuring a curved aluminium body and classic click wheel has reignited nostalgia as refurbished iPods surge in popularity, despite it being unlikely that Apple will ever revive the product.

A PEEK INTO REDDIT'S ANTI-SPAM INTERNALS

This article is an in-depth analysis of Reddit's anti-spam mechanisms, showing how moderation tools and processes work, including the use of various algorithms like spamurai and the Perspective API. An unexpected glitch showed internal reasons for post removals, allowing insights into the criteria and methods used by Reddit to detect and eliminate spam.

QUALITY ASSURANCE AGENT: REIMAGINING SOFTWARE QUALITY WITH AI-DRIVEN AUTONOMOUS TESTING

LinkedIn built an autonomous QA Agent using AI and vision-language models to test its apps across the huge range of UI permutations that manual and scripted testing can't keep up with. It runs a multi-model setup (planning, error detection, and a fine-tuned model that converts instructions into screen coordinates) and a hybrid “System 1/System 2” architecture: fast deterministic replay when the UI is unchanged, and slower agentic reasoning when it shifts.

HOW WE REALLY BUILD PRODUCTION-GRADE AI AGENTS: BEYOND MODELS, TOWARD DATA AND API QUALITY

Production-grade AI agents fail mostly from interface problems, such as underspecified APIs, ambiguous data, and missing guardrails. Teams should stop over-indexing on model intelligence and treat agents as distributed systems built on data quality, API quality, and execution quality. “Agent-ready” APIs help agent quality improve overall.

AI SHOULDN'T SHRINK HEADCOUNT. IT SHOULD SHRINK TEAMS

AI helps smaller teams by allowing them to make decisions faster, have clearer ownership, and better visibility of progress.

IS AI MAKING YOUR TEAMS BETTER, OR JUST BUSIER?

AI adoption often leads to increased tool usage without fundamentally improving team capabilities, resulting in faster completion of the same tasks rather than new achievements. Good measurement of AI impact focuses on two key performance indicators: whether AI enables outcomes that weren't previously possible and how thoroughly AI is integrated into daily workflows.

WHY SPECIALIZATION IS INEVITABLE

Specialization is a necessary principle in AI, where systems perform best when narrowly focused on specific tasks rather than trying to be general-purpose. Mathematical and empirical evidence across optimization theory, biology, and competitive markets consistently shows that concentrated effort leads to better performance in constrained environments.

INTRODUCING CLAUDE SONNET 5

Claude Sonnet 5 is a new AI model that improves agentic capabilities, allowing it to autonomously perform tasks such as coding and browsing, while offering better performance and safety compared to its predecessor, Sonnet 4.6.

EXPO 57

Expo SDK 57 has been released, featuring an upgrade to React Native 0.86, which is designed to be a straightforward upgrade with no breaking changes. This release aims to shift towards a new Expo SDK release cadence that allows for simpler, non-breaking updates between major releases, enhancing the experience of app developers.

GITHUB STACKED PRS

Stacked PRs allows users to break large changes into smaller, manageable pull requests that build upon each other. With both GitHub UI and a powerful CLI, this feature allows for seamless navigation, status checks, and cascading rebases across multiple PRs.

I PORTED KUBERNETES TO THE BROWSER

A new project called webernetes has been developed as a partial port of Kubernetes to TypeScript, allowing users to run Kubernetes clusters directly in their browsers. The implementation includes features such as pod lifecycles, deployment tracking, and a simulated networking environment.

APPLE FIXES WEBKIT FLAWS IN IOS AND MACOS, WITH HELP FROM AI TOOLS

Apple has released security updates for iOS, iPadOS, macOS, and Safari, addressing four WebKit vulnerabilities discovered with the help of AI tools. The company expedited these fixes due to the pace at which AI can convert known flaws into working exploits.

you through a run. Read the blog

DON'T MAKE GATES OPTIONAL, MAKE THEM FLEXIBLE

Implementing a required but flexible approval process allows teams to take calculated risks while maintaining speed and confidence.

WORSE IS BETTER

“Worse is better” suggests that sometimes inferior technologies or practices gain popularity over seemingly better alternatives, but this is mostly because “perfect” is the enemy of good.

MICROSOFT NEEDS WINDOWS LITE

Microsoft should create Windows Lite, a stripped-down version of Windows without telemetry or ads, to retain developers and gamers.

THE BEST THING THAT'S EVER HAPPENED FOR MULTIPLAYER GAMES?

Amazon GameLift Servers now offers free network bandwidth for multiplayer games hosted on its platform, reducing hosting costs for indie developers and smaller teams.

LET IT CRASH: HOW TO STEER WHAT COMES AFTER

An AI crash could still leave behind power plants, data centers, open models, and a workforce trained to work with machines. Overbuilding is how new infrastructure usually gets installed. The crash only opens the door - the rebuild depends on who already owns the compute, models, and tools when prices reset. Spread the means, not the proceeds.

ONTOLOGY EVERYWHERE!

An ontology is a formal, explicit specification of a shared conceptualization. It is an agreement about what things exist in a domain, how they relate, and what we call them. As AI agents become the primary consumers of data, we need a layer that explicitly describes what things mean, how they relate, and what rules govern them. Structure knowledge and make it accessible, and the market will sort out the labels.

HIRING WITH AI

Recruiting feels perfectly aligned with AI capabilities because, at its core, it's mostly a data problem. This article looks at the state of the AI recruiting industry. Every AI recruiting company seems to run into the same problems: recruiting isn't very frequent, causing natural user churn; companies can use multiple platforms at once, so there's no loyalty or lock-in; a data moat is hard; and the long-term winner needs to become part of the broader recruiting window to scale. The space is still early, and markets can be unlocked with new capabilities.

WHAT IS TALENT AND TALENT ENGINEERING

'Talent' and 'Talent engineering' are two new roles that have recently emerged. Frontier labs, VCs, and startups are hiring for both roles. A 'talent' is a person who is great at attracting smart people and has deep relationships with all of them. A 'talent engineer' is someone who loves to build people tools and is also a great engineer. The highest talent density companies now hire based on a blend of talent and talent engineering rather than using the traditional model of recruitment.

HOW TO KILL CHURN

You can't out-onboard a customer who only needed you once. 10% monthly churn means rebuilding 70% of your user base every year. The biggest lever is category choice, because a daily marketer workflow retains nothing like a one-time birthday-card job. Blended churn hides the split between sticky cohorts and tourists, so the work is steering ICP, onboarding, and acquisition toward the people who come back. Annual plans, dunning, and win-back flows help at the margin, but choosing a repeat-usage customer is the whole game.

WHAT GAME ARE YOU PLAYING?

A lot of founders are describing their company with the wrong game board. PLG, enterprise trust, marketplace liquidity, and liquidity financing each have different rules, clocks, and ways to die. Name the win condition first, then the feedback loops and constraints that follow from it. Strategy gets sharp the moment you can name the game everyone else is misplaying.

X NOW OFFERS AN MCP SERVER TO MAKE ITS PLATFORM EASIER FOR AI TOOLS TO USE

X has released a Model Context Protocol (MCP) server that lets AI tools communicate with the X API. The new MCP doesn't add new capabilities beyond what X's API already offers - it just makes them easier to use for AI applications. The integration turns X into an information network filled with real-time data to retrieve and analyze. The MCP tool is not compatible with X's Write API endpoints, so it can't be used to post autonomously.

INTRODUCING CLAUDE SONNET 5

Claude Sonnet 5 can make plans, use tools, and run autonomously for long periods. Its performance is close to that of Opus 4.8 but at lower prices. Safety assessments show that Sonnet 5 shows a lower rate of undesirable behaviors than Sonnet 4.6 and is generally safer to use in agentic contexts. It has a much lower ability to perform cybersecurity tasks than current Opus models. Claude Sonnet 5 is now available across all plans.

NANO BANANA 2 LITE

Nano Banana 2 Light, also known as Gemini 3.1 Flash Lite Image, is the fastest and cheapest Gemini image model. The model produces better results than previous generations, but it still makes some spelling mistakes. This post shows what it produced when prompted for a Where's Waldo-style image.

THE PRIMITIVE IS THE PRODUCT

AI has created a new type of user that is rapidly consuming an expanding share of software. Agents intuit and compose software instead of navigating it. They care purely about capabilities and whether they can be invoked and chained reliably. The best product for an agent is just the best product for a developer - software companies need to start thinking like developer tools companies.

THE WORLD AS MODEL

High-frequency trading firms succeeded by developing systems that effectively extracted predictive signals from market data at millisecond timescales. Companies are now applying similar approaches to the physical world, creating digital representations of industrial processes to optimize financing and trading of physical assets. The focus is on developing the "reality stack" to turn physical signals into actionable financial intelligence, reshaping markets and investment strategies.

BEYOND P(DOOM): MARC ANDREESSEN - BETTING ON AMERICA (65 MINUTE PODCAST)

Marc Andreessen argues that the primary obstacles to AI-driven productivity are institutional and infrastructure-based rather than technological.

THE FOUNDER'S GUIDE TO SECONDARIES: GETTING LIQUID BEFORE THE EXIT

While liquidity is heavily concentrated among a few top-tier names, structured company-run tender offers are increasingly utilized to distribute capital on a recurring basis without disrupting 409A valuations.

I AM OBSESSED WITH THESE TWO ADS

An Omnilux LED face mask and Dyson's "AI-powered British strawberries" explain more about growth than another prompt hack.

BREAKING DOWN RETAIL MEDIA'S VIBE SHIFT

Amazon and Walmart now control 85% of the retail media market. Walmart's $1.4B acquisition of Vibe widens that gap further. The deal mirrors Amazon's playbook of plugging first-party commerce data into a DSP and striking deals with premium streamers. Walmart has 200,000 Marketplace resellers it can immediately convert into CTV advertisers. For the 200-plus other retail media networks holding the remaining 15% between them, media planners aren't allocating budget to a third network after maxing out the top two, which makes data cooperatives or acquisition the only realistic paths forward.

EVERYTHING IN MARKETING STILL WORKS. IT JUST DOESN'T DIFFERENTIATE YOU ANYMORE

AI flattened execution, so channels, volume, and best practices stopped signaling anything. The only moat left is craft. Taste, a sharp point of view, and the judgment to go deep on one channel instead of spreading thin. The work that drives results now is the least measurable. Brand, thought leadership, and owned media that compound over time. It's slower, harder to prove, and often looks wrong on a dashboard, but it's what actually moves buyers.

WHY PROPRIETARY DATA IS YOUR MOST DEFENSIBLE AI CITATION ASSET

Publishing proprietary first-party data gives content a stronger chance to earn AI citations and organic visibility, but page structure often decides who gets the citation. One study found pages with 15 or more unique data points averaged an information gain score of 62.1 versus 40.2 for pages with one or fewer. Put your strongest statistic, methodology, and key findings in the first 30% of the page because 44.2% of ChatGPT citations come from that section. Build comprehensive pages that answer multiple related questions and make original data easy for AI to extract instead of hiding it in long narratives.

FIND WHAT BUYERS ASK AI

With AI Overview traffic now harder to track, stop chasing clicks and start understanding what buyers actually ask AI. This post includes a templated prompt to generate real customer questions, combine the results, and then analyze overlaps to find the most consistent themes. Those repeated questions reveal what matters most to your audience, giving you high-signal ideas for offers and campaigns.

SOCIAL MEDIA SEO WORKBOOK

Social content now needs to be built for discovery. This playbook shows how to do it with keyword research, search intent, content planning, and profile optimization, plus tactics like using native search data, matching formats, and leaning into engagement signals. It also includes a free downloadable social media SEO content planner.

START BUILDING WITH NANO BANANA 2 LITE AND GEMINI OMNI FLASH

Google just dropped two new generative media models built for creative iteration. Nano Banana 2 Lite is its fastest and cheapest image model yet, optimized for rapid, high-volume image generation. Gemini Omni Flash brings high-quality video generation and conversational editing using text, image, and video inputs. Together, they let users quickly go from idea to image to video in one workflow. Both are available across Google AI Studio, APIs, and consumer products, with competitive pricing and a few early limitations on video length and consistency.

BAD DATA MEANS BAD DECISIONS

Relying on one attribution model leads to bad marketing decisions because it cannot capture every source of revenue. One brand saw reported conversions fall after it increased video investment because first click attribution could not track view through impact. Post-click conversions fell from about 80% to 20% even though incrementality testing showed revenue increased. Compare incrementality studies, media mix modeling, and platform data instead of treating one reporting system as the source of truth.

WHY CARE ABOUT THE STUPID EM DASH ANYWAY?

Fear of looking AI-generated has created a form of reputational self-surveillance, where marketing teams are removing normal punctuation to preempt accusations that have not been made. The concern behind the trend is real as AI leans on certain marks and constructions hard enough that they start to feel like tells. The response should not be banning punctuation - it should be building a writing standard specific enough that the copy is unmistakably yours.

HOW AI MENTIONS AND CITES YOUR BRAND

Only 23.1% of brand mentions in AI responses include a citation to the brand's domain, meaning AI covers brands at a much higher rate than it sources their content. A study of 12,000 responses across Google AI Mode, AI Overviews, Gemini, and GPT-5.4-mini found that prompt type drives the biggest gap. Single-brand queries earn a 39% mention-to-citation rate, while open-ended category queries drop to 7.2%. Missing citations tend to signal an SEO and GEO problem, since AI can't find content worthy of sourcing. Missing mentions tend to signal a PR problem, since the brand isn't present enough in third-party training data.

WHERE TWO METRICS MEET, THE DIAGNOSIS APPEARS

Pair metrics like AOV and conversion rate to get a more accurate picture of results, compared to looking at one metric alone.