Search the archive

POMELLI

Pomelli is an experimental AI-powered marketing tool from Google Labs and DeepMind that helps small businesses create on-brand social media posts, ads, and campaign visuals automatically by analyzing their website.

SCREEN RECORDINGS THAT STAND OUT

Create beautiful screen recordings for product demos, tutorials, and explainers.

YOUR ALL-IN-ONE AI MUSIC STUDIO

AI vocals, AI instruments, voice cloning, stem splitter, music generator, and more—all in one place.

ARE YOU DESIGNING FOR THE USER'S VALUES — OR YOUR OWN?

As technology becomes more pervasive, the designer's role is shifting from crafting interfaces to shaping ethical guardrails for how people interact with machines. Empathy and traditional UX methods don't capture user values, so designers often unknowingly impose their own. Explicit ethical frameworks and evaluation tools are needed to surface value trade-offs—such as inclusion, autonomy, transparency, privacy, and well-being—and ensure products consciously align with users' values rather than the team's assumptions.

GOOGLE IS BUILDING AN EXPERIMENTAL NEW BROWSER AND A NEW KIND OF WEB APP

Google's Disco is a new AI-powered browser that uses Gemini to turn searches into “GenTabs”—interactive, task-specific mini apps built from both AI suggestions and the web pages users open themselves. Launched in Google Labs, Disco explores a more collaborative, web-forward model where browsing, research, and on-the-fly app creation merge. Google is still unsure whether GenTabs should be temporary, shareable, or evolve into a broader product.

DESIGNING EFFECTIVE DASHBOARDS UX WITH ACTION DOTS

Action dots are visual indicators that help dashboard users immediately identify which metrics need attention by categorizing values into five threshold states. Unlike traditional dashboard indicators, which often create false positives or bury problems, action dots appear only when metrics fall outside the neutral zone, making critical issues instantly visible while letting non-urgent data fade into the background. This approach addresses common dashboard failures, such as information overload and unclear action priorities, by focusing attention on what genuinely requires user intervention.

WHY RETRO, INNOVATIVE INDIES AND NINTENDO RULED GAME DESIGN IN 2025

2025 proved a strong year for indie and AA-budget games like Clair Obscur: Expedition 33 and Hollow Knight: Silksong, which won design plaudits over major publishers' offerings.

EIGHT CRITICAL USABILITY TESTING MISTAKES AND HOW TO AVOID THEM

User testing reveals usability problems by observing real people interact with websites or apps, but many UX teams make critical mistakes that distort results and waste resources.

3️⃣ Tooling evaluation framework + scorecard

Tackle problems early in the lifecycle by accelerating your

prototyping strategy. Get the guide to learn how.

CLAUDE AGENT SKILLS: A FIRST PRINCIPLES DEEP DIVE

MAKING COMPLEX WEB APPS FASTER

Microsoft is proposing a new Delayed Message Timing API to help web developers diagnose and fix performance issues in complex web applications that use multiple windows, iframes, or worker threads. The API addresses three main causes of messaging delays: busy receiving contexts, congested task queues, and serialization/deserialization overhead by providing detailed timing metrics and diagnostic information.

ASK HN: HOW CAN I GET BETTER AT USING AI FOR PROGRAMMING? (HACKER NEWS THREAD)

A dev is struggling to effectively use AI, specifically Claude Code, to rewrite an old jQuery and Django project into SvelteKit. Experienced users recommend using a CLAUDE.md file correctly for persistent instructions, using Claude's "plan mode" for task breakdown before implementation, and adding feedback loops through tests to verify the AI's work. Also, Opus 4.5 should be used over any other model for coding when possible.

JUSTIFIED

just can normalize most projects in a convenient, fast, readable, and portable way. Large language models are good at generating justfiles, and justfiles help models know what actions they can run on a project. This post goes through some tricks you can do with just.

JSDOC *IS* TYPESCRIPT

Svelte's move to JSDoc for type declarations was not a rejection of TypeScript. Rather, JSDoc should be understood as an inherent part of TypeScript itself. This is because the TypeScript language service powers the static analysis and IntelliSense for both `.ts` files and JSDoc comments, making them two sides of the same coin.

SHADCN/CREATE

shadcn/create lets developers build their own shadcn/ui. Everything is customizable - developers can change components, icons, colors, themes, and fonts to build something unique. shadcn/create is available for Next.js, Vite, TanStack Start, and v0. A video introducing the framework is available in the thread.

CAPSUDO

capsudo is an object-capability style sudo. It creates a socket that acts as an object capability, allowing anyone who can access the socket to make use of it. The capsudo daemon accepts connections, stitching everything together to run programs bound to object capabilities. Examples of how to use the tool are available in the repository.

THE UNOFFICIAL 37SIGNALS/DHH RAILS STYLE GUIDE

This GitHub gist contains a Claude Code-generated style guide based on a deep analysis of 37signal's open source project management tool, Fizzy. Claude Code was used to analyze the entire code base to extract the patterns used and infer the philosophy of implementation choices. The guide contains patterns extracted from actual code, guidance based on observed conventions, notes on what's deliberately absent, and reusable code that can be copied into personal projects.

CODING AGENTS & COMPLEXITY BUDGETS

Cursor migrated from a headless CMS to raw code and Markdown after realizing the CMS created unnecessary abstraction and slowed down content updates. They used Cursor's AI coding agents and completed the migration in just three days, spending only $260 in tokens to automate content export, conversion, and codebase refactoring.

YOU SHOULD NEVER BUILD A CMS

The Cursor migration from Sanity to a markdown and Git-based system mentioned above was responded to by Sanity. Sanity acknowledged valid frustrations with clunky preview workflows, authentication fragmentation, and AI agent accessibility. However, they still found that Cursor's new system still has trouble at scale due to issues like content denormalization, poor collaboration, and limited querying capabilities for AI.

HOW GOOD ENGINEERS WRITE BAD CODE AT BIG COMPANIES

Big tech companies produce surprisingly sloppy code because most code changes are made by engineers who are relative beginners to the specific codebase, programming language, or system they're working on. This happens due to high employee turnover, frequent internal reorganizations, and compensation structures that incentivize job-hopping.

WHAT IS A BUILD SYSTEM, ANYWAY?

A build system is pretty much anything that lets you specify dependencies on a previous artifact.

HOW TO LOAD CSS (FAST)

Compression dictionaries have the potential to revolutionize the way CSS is served to users, as they significantly reduce costs.

ENGINEERING INTELLIGENCE

Intelligence is engineered, not trained: models are components, systems create value, and engineering makes machine learning real.

SURGERY ON CHROMIUM SOURCE CODE: REPLACING DEVTOOLS' HTTP HANDLER WITH REDIS PUB/SUB

For reliable and scalable automation of numerous Chromium browser sessions, this dev directly integrated Redis Pub/Sub messaging into Chromium's DevTools pipe handler to remove the instability and complexity of traditional TCP/IP communication.

I FED 24 YEARS OF MY BLOG POSTS TO A MARKOV MODEL

This dev showed off his simple Markov text generator, Mark V. Shaney Junior, showing the "gibberish" it produces after being trained on various texts.

USE PYTHON FOR SCRIPTING!

Python is great for scripting due to its widespread availability, standardized library, and clearer syntax.

HOME DEPOT EXPOSED ACCESS TO INTERNAL SYSTEMS FOR A YEAR, SAYS RESEARCHER

Home Depot's internal systems were exposed for a year due to an employee's mistakenly published access token, which a security researcher discovered and reported, but the company ignored the alerts until TechCrunch intervened to prompt a fix.

Watch the recording

LONG LIVE SYSTEMS OF RECORD

Everyone keeps saying AI agents will kill systems of record. The opposite might be true. Ask sales what your ARR is, and you get one number. Ask finance, and you get another. Humans can talk it out - agents can't. The more we automate, the more someone needs to decide what the correct answer actually is.

THE REAL STATE OF SEED TODAY: THE TOP 10 LEARNINGS FROM 50,000 STARTUPS, PER CARTA'S LATEST DATA

The seed market is healthy but bifurcated. AI companies are operating in a different universe than everyone else, who is competing for a smaller share of attention with leaner teams and longer runways. SAFEs have become completely standardized, and solo founding is increasingly viable, but remains a handicap for fundraising. The fundamentals are still the same: build something people want, get to metrics that matter, and raise when you have the leverage.

THE AI VALUE GAP: WHERE DOES THE $7,000 PER SEAT GO?

OpenAI claims that AI saves the average white-collar worker 54 minutes per day. Currently, vendors capture 10% to 15% of that value, leaving employers and employees with 85% to 90% value capture. This gap suggests that significant pricing power remains untapped. Bundling further complicates the picture - there appears to be a market for best-in-class vertical tools even when bundled alternatives exist.

THE AGE OF COMPOSABLE SOFTWARE

Software has always faced a brutal trade-off - you can have customization or affordability, never both. Affordability always won. That's why personal computers were never really personal, and most apps follow "one codebase to rule them all." AI changes this math so dramatically that the trade-off might finally break down. Instead of killing software, AI could make it infinitely personalized - built for an end-market of one.

PLAY RIGGED GAMES

Unfair advantages come in three flavors: product insight, go-to-market edge, or technical moats. One common thread among successful founders is an obsession that rational people would walk away from. They solved problems they couldn't stop thinking about. Find your edge and stop fighting fair.

HOW MUCH EQUITY CAN YOU GIVE TO EMPLOYEES BEFORE IT'S A PROBLEM?

The median public tech company spends 12% of revenue on stock-based compensation. Sounds outrageous until you see that actual shareholder dilution is under 2%. The math is simple - SBC is measured against revenue, dilution against market cap. A company trading at 4x revenue spending 12% on SBC only dilutes 3% of market cap. The companies getting killed aren't spending 20% on SBC. They're the ones growing 5% while pretending they're still a growth story.

equation. Learn more about white-label payments with NMI

MACALY

AI platform for creating and managing websites and apps with built-in editing, data tools, and publishing.

COMMITIFY.ME

AI voice agent that calls your phone like a real person to keep you accountable, focused, and on track—no app required.

CLEARITTY

AI-powered sales intelligence with human-verified buyer signals and real-time intent scoring to target only in-market leads and close deals faster.

AI PULLS VC AND PE INTO THE SAME ORBIT

AI is transforming labor-heavy service sectors, putting SMB and mid-market businesses within reach for tech startups and collapsing the historical divide between venture capital and private equity. PE funds are becoming distribution channels for AI products, while founders use PE portfolio pages as idea maps as AI expands TAM for legacy software categories. AI native rollups accelerate the convergence as VC-backed platforms acquire traditional services firms and rebuild them around automated operations.

THE MAP IS NOT THE TERRITORY

VCs treat the power law like a selection tool when it's actually just a description of what happens after the dust settles. They look at seed-stage startups and ask, "Is this a power law company?" - but that's a category error. Uber looked like a black car service for rich people. Airbnb was for travelers who couldn't afford hotels. Both got passed because investors couldn't model a $10B outcome. The irony is that power law returns almost always come from market creation, not capturing predictable share of existing markets.

ESSENTIAL AI INTEGRATION BEST PRACTICES FOR OPTIMAL PERFORMANCE

Implement migration patterns in AI API calls to maintain flexibility and reliability amid frequent model changes. Utilize service tiers like OpenAI's Flex for cost-efficient backend operations, cutting expenses by up to 50% without sacrificing data quality. Employ rate limiting and circuit breakers to manage resource usage effectively and mitigate unexpected costs or abuse in AI systems.

No trial, no credit card - it's free. Check it out

SIX BIG BETS

You don't have to take more risk, just structure your life to be exposed to the upside.