OpenAI shut down its Sora video app and developer APIs, citing unsustainable compute costs, low user retention, and legal liabilities, pivoting instead to enterprise software and coding agents. xAI is moving in the opposite direction, leveraging its built-in distribution through X and financial backing from SpaceX — which merged with xAI in a $1.25 trillion deal — to absorb the high inference costs that sank Sora. Unlike OpenAI's standalone app approach, xAI's Grok Imagine can be embedded natively into X's feed, where short-form video content already thrives.
In an era where AI makes creating fast and effortless, the key differentiator is no longer production but taste—the learned ability to judge what is meaningful, appropriate, and worth making. Drawing on thinkers like David Hume and Pierre Bourdieu, taste is framed as a discipline developed through exposure, time, and critical reflection—especially during the often-skipped “incubation” phase of creativity—without which output becomes shallow and repetitive. As a result, value now lies in discernment, context, and judgment, which cannot be automated or rushed.
AI represents "the Creative Infinite" - an unprecedented ability to have something exist within minutes of its description. This technology has completely eroded traditional barriers to creation, allowing people to translate their natural thoughts directly into creative outputs. Rather than replacing existing creative tools, AI compounds with them to exponentially expand human creative potential in ways we haven't fully grasped yet.
Agentation is a tool that turns UI annotations into structured context for AI coding agents by letting users click elements, add notes, and generate formatted output with CSS selectors and component paths.
Sunday's Memo is a home robot designed to automate everyday tasks using AI that learns directly from human demonstrations via its Skill Capture Glove. Working with Moniker, the company built a brand focused on simplicity and trust—balancing advanced technology with a calm, approachable identity—emphasizing real-life usefulness over hype, which helped drive strong early interest, rapid waitlist growth, and major investment.
AI is increasingly being used in product operations to automate repetitive tasks, improve decision-making, and free up teams for higher-impact work. Technologies ranging from generative AI and RPA to predictive analytics and conversational chatbots can be applied across data analysis, workflow automation, and cross-team communication. While the benefits include faster task completion and scalable operations, organizations must also navigate risks such as data privacy concerns, AI model bias, and change management challenges.
The Moxy Kuala Lumpur Chinatown hotel aligns with the city's diverse cultural and religious landscape, where mosques, temples, and modern towers coexist in parallel rather than in conflict.
Wacom has announced an upgrade to its Bridge technology, enabling artists to draw anywhere by connecting their tablets to powerful cloud-based computers.
A version of DOOM was developed that uses CSS for all visual rendering, including 3D projection and lighting, while JavaScript handles the game logic. It shows how modern CSS features like `transform` and `@property` can manage complex geometry and animations without relying on traditional rendering engines.
Colossus, Google's foundational distributed storage system, supports all Google products with high throughput, scalability, and various performance requirements. It gets SSD performance at HDD prices by using L4, an advanced distributed SSD caching technology that uses machine learning to optimize data placement for frequently accessed "hot" data.
The author decrypted Cloudflare Turnstile programs used by ChatGPT, revealing a sophisticated bot detection mechanism that analyzes the internal state of the React application to make sure it's a genuine environment. This system identifies bots by verifying application-level state and behavioral biometrics, although its security relies more on complex obfuscation than true encryption.
The author reflects on 40 months of AI, acknowledging the utility of interfaces like Claude Code while questioning AI's overall impact on coding productivity and creative quality. Despite recognizing its potential for expert-level tasks, he is still skeptical of its ability to replicate personal style or handle complex software development without human intervention.
NOT REGISTERED YET? Final call! Sign up before April 17th to compete for a share of $180K [10] while helping solve the toughest industrial manufacturing problems. No electronics expertise required. Developers, roboticists, and engineers, get ready to apply your skills in this global challenge brought to us by Intrinsic, Open Robotics, NVIDIA, and Google DeepMind. REGISTER NOW [10]!
Pretext is a pure JavaScript/TypeScript library for fast and accurate multiline text measurement and layout. It circumvents costly DOM reflows by implementing its own text measurement logic using the browser's font engine. The library has two primary use cases: calculating a paragraph's height without DOM interaction for purposes like virtualization, and providing tools to manually lay out text lines for rendering to Canvas, SVG, or server-side.
AI agents are already causing data loss, from wiped home directories to deleted files, when granted ordinary machine access. jai provides an effortless, one-command solution for containing these agents on Linux, bridging the gap between full system access and complex virtualization. It works by giving your working directory full access while protecting your home directory with a copy-on-write overlay or complete isolation.
The early internet was an open "bright meadow" where sharing ideas and code led to innovation and success. However, corporate consolidation and the advent of AI have transformed it into a "cognitive dark forest." In this new environment, AI platforms can cheaply absorb emerging human innovations by analyzing user interactions and data, effectively owning the compute, models, and developer data.
After exhausting standard treatment options for his osteosarcoma, GitLab co-founder Sid took agency to develop and scale personalized diagnostics and parallel treatments for himself and others using AI.
Merrilin is a reading app that uses a dual-lane synchronization system, combining peer-to-peer and cloud-based methods, to manage multi-device progress and annotations through distributed systems principles, which are discussed in this article.
While debounce smooths UI input, it fails to manage network request lifecycles, requiring cancellation, retries, and proper error handling for truly reliable applications.
Most tools are still locked to their own database, blind to everything users already have in Slack, GitHub, Salesforce, Google Drive, and dozens of other apps. That's the ceiling on what you can build.
190 companies presented this week. 85% were AI-first. 14 hit $1M ARR before demo day, the highest ever for any YC batch. The bigger signal here is that the bar for what counts as "early stage" keeps moving. $1M ARR used to be a Series A milestone. Now, it's table stakes for a demo day presentation.
The AI market isn't the same as the Dotcom bubble of the past. The technology is still early, and demand is pulling spending forward. The software selloff is real, but it won't hit every layer equally, as each has its own moats. Across the last few technological revolutions, the companies that became durable winners were predominantly founded in years four and five of each platform shift. We are now in that window for AI.
Every major tech shift creates a new growth channel. AI is becoming a decision-maker - they're starting to choose products. This new growth channel is embedding products directly into the AI tools that people use. This impacts strategy, product, and engineering. The companies that will be marketing to the interface layers that make decisions on users' behalf.
2026.03.30·
mattgiustwilliamson.substack.com·6 min read·TLDR Founders
The hardest problem in AI right now isn't the model, it's the pricing. In traditional SaaS, you could underprice a seat, and it didn't matter because the marginal cost of serving another user is basically zero. Now every agent action, every "magic" feature triggers real inference costs on the backend. Price too low and you're literally paying customers to use your product. Price too high and you lose to the competitor who figured it out first. Per-token and credit-based pricing models are about to become the default because they're the only ones that align cost structure with value delivered.
2026.03.30·
cloudedjudgement.substack.com·8 min read·TLDR Founders
Antifragility is the premise that some things, including people, improve from stressful situations. Disorder and uncertainty are the rule at startups. Over time, antifragile people will consistently improve their value/skills due to the continued exposure to a startup's disorder.
Software companies have three paths right now: accelerate revenue 10%+ YoY through AI-native products, get stuck in the messy middle, or rebuild to 40-50% operating margins. Even the profitability path is deceptive. Most companies that look profitable on FCF margins are still in the messy middle once you account for stock-based compensation. The only real option is to accelerate. If you're not rewriting your product around AI, you're choosing a slow death.
ChatGPT generated over $100M in annualized ad revenue just 6 weeks after launching its ad pilot program. Less than 20% of eligible US users currently see daily ads. The platform now hosts more than 600 advertisers and plans to open self-serve access in April, potentially expanding to Canada, Australia, and New Zealand. Since around 85% of Free and Go tier users are eligible for ads, the current revenue represents only a fraction of the platform's total advertising capacity.
Reddit is tightening control over automation and giving an advantage to brands that rely on real human engagement. Starting March 31, the platform will label automated accounts, add human verification for suspicious activity, and expand reporting tools. Reddit is already removing about 100K spam accounts per day. This reduces the effectiveness of bots and exposes brands that rely on automation or fake accounts. Brands should register approved automation before the June deadline and invest in real human presence to stay competitive.
Clarify removes CRM seat fees and charges only when its AI performs useful work, which aligns cost directly to value. Traditional CRM setups can take 3-6 months to deliver value and can cost $10K per month from the start, even before adoption. It solves messy “Frankenstack” systems by combining relational, time series, and unstructured data into one unified customer record that AI can act on. Its credit-based pricing removes adoption friction and ensures customers pay only when outcomes like deal updates or meeting insights happen. This model also supports a compound startup strategy where new features require about 20% new work and build on an existing data and AI foundation.
The spray-and-pray approach to outbound is failing, but teams that treat it as a deliberate, focused motion are still making it work. Top performers are capping account books at around 75, clustering calls, emails, executive LinkedIn messages, and physical mail into short coordinated sprints, and using AI to surface how customers actually describe their problems rather than to mass-produce emails. Infrastructure matters too, as domain reputation and bounce rates now determine whether messaging strategy even gets a chance to land.