Search the archive

THE PODCAST WHERE YOU CAN EAVESDROP ON THE AI ELITE

Dwarkesh Patel hosts a podcast that averages two million listens per episode featuring guests like Jensen Huang, Mark Zuckerberg, and the most influential AI researchers.

See how today's cloud attacks work

GOOGLE WILL INVEST AS MUCH AS $40 BILLION IN ANTHROPIC

Google will invest between $10 billion and $40 billion in Anthropic. The amount depends on whether Anthropic can meet certain performance targets. Anthropic recently received a $5 billion investment from Amazon, with an option for more investment based on performance. The investments value Anthropic at $350 billion. The funds will help the startup close the gap between demand and supply of compute for AI training and inference.

WHAT HAPPENS WHEN AI RUNS A STORE IN SAN FRANCISCO?

Andon Labs is running an experiment to see whether AI agents can run real-world endeavors. It opened a retail boutique on April 10 run by an agent named Luna. Luna has so far struggled with employee schedules and seems to be unable to stop ordering candles. The experiment's mission was to make a profit, but it has lost $13,000 since the shop's opening.

ANTHROPIC LAUNCHES MEMORY IN CLAUDE AGENTS FOR ENTERPRISE

Anthropic has released a feature for Claude Managed Agents called Memory. It allows agents to remember and use information from prior sessions and accumulate knowledge over time without requiring manual prompt updates. Memory is a filesystem-based layer, so data is stored as files that can be exported, managed through APIs, and scoped with permissions for various organizational needs. The feature is available now in public beta to all Managed Agents users.

GOOGLE PREPARES CREDITS SYSTEM FOR GEMINI

Google is working on a credit-based system for its Gemini app where users receive a monthly allowance to spend across models and features. Users will be able to top up when they run out of credits. The change will make budgeting for heavy workloads more predictable and give Google a cleaner lever for introducing premium features without forcing users to pay for more expensive plans. OpenAI, Anthropic, and Notion already use a similar consumption model.

YOUR AI MIGHT BE LYING TO YOUR BOSS

It's very hard to measure the contribution that AI models make to a codebase. Sometimes the best use cases for AI are inquisitive prompts that don't necessarily produce any code at all. Lines of code isn't a very good measure of code quality, and it can be difficult to separate the work engineers did vs what AI has done. The bias appears to be towards reporting a higher AI percentage, which is great for AI companies, but skewed metrics can be harmful.

THE WORLD CAN'T KEEP UP WITH AI LABS

Coding agents are the first AI product people are paying for at volume and regularly. However, compute demand has started to grow faster than anyone can build it out. The industry isn't ready for the agent boom. The most obvious move for AI labs now is to cut limits and raise prices.

MONITORING LLM BEHAVIOR: DRIFT, RETRIES, AND REFUSAL PATTERNS

Monitoring LLM behavior necessitates adopting the AI Evaluation Stack, separating tests into deterministic assertions (syntax and routing integrity) and model-based evaluations (semantic quality). Engineers use offline pipelines for pre-deployment regression testing with human-reviewed "Golden Datasets" while online pipelines monitor real-world performance for drift and failures. A continuous feedback loop from production telemetry ensures AI systems adapt, maintaining high performance as user behavior evolves.

see it in action

STASH

Stash is a tool that gives agents persistent memory. It enables agents to remember, recall, consolidate memories, and learn across sessions. Stash is open source, self-hosted, and works with any MCP-compatible agent.

VISION BANANA GENERALIST MODEL

Instruction-tuned image generation models can act as generalist vision systems, achieving state-of-the-art results across tasks by reframing perception as image generation.

EFFICIENT VIDEO INTELLIGENCE IN 2026

Efficient video intelligence advances include compact universal vision encoders like EUPE, which distill capabilities from specialized models such as DINO and SAM. Techniques like LongVU use adaptive token allocation and compression for long-form video understanding while edge and on-device deployment handle real-time processing. Persistent challenges include streaming understanding, sparse-event detection, real-time sub-watt inference for AR glasses, and robust multi-modal reasoning.

SCALING LONG-HORIZON CODING AGENTS

A framework from Meta for test-time scaling in coding agents summarized past rollouts into structured representations, enabling better selection and reuse to improve benchmark performance.

OPENAI POSTS FIVE-PRINCIPLE FRAMEWORK FOR AGI, ALTMAN CONCEDES BIGGER ROLE

OpenAI has published a five-principle framework for the development of artificial general intelligence. It is the company's most prominent statement of intent since its 2018 Charter. The lab claims it will resist letting the technology consolidate power in the hands of the few. The framework arrives at a time when US and European regulators are tightening oversight of frontier AI labs.

CURSOR'S $60 BILLION ESCAPE HATCH

What does it mean when a company doing $2.7B in annualized revenue has gross margins of negative 23%? In Cursor's case, it means AI coding tools have inverted the old SaaS playbook, where serving the next customer is supposed to be cheap. Power users consume more model capacity and compute, so the best customers can become the most expensive. That reframes the rumored SpaceX deal as more than a $60B headline. Access to Colossus would loosen Cursor's dependence on Anthropic and OpenAI fees, where that negative 23% lives.

META'S LOSS IS THINKING MACHINES' GAIN

Thinking Machines Lab has been hiring more researchers from Meta than from any other single employer. The AI startup is expanding on multiple fronts, and it just signed a multibillion-dollar cloud deal with Google that gives it access to Nvidia's latest GB300 chips. Meta's large pay packages are well known, but Thinking Machines' $12 billion valuation and 140-employee count mean there's still a lot of financial upside to joining the startup.

BECOME A CURATOR FOR TLDR AI (3-5 HRS/WEEK, FULLY REMOTE)

We're looking for an engineer/researcher at a major AI lab or startup to help write for 1M+ subscribers. Curators have been invited to Google I/O and OpenAI DevDay, scouted for Tier 1 VCs, and get early access to unreleased TLDR products. Learn more [21].

AN AMATEUR JUST SOLVED A 60-YEAR-OLD MATH PROBLEM—BY ASKING AI

The raw output of ChatGPT's proof was quite poor and required experts to sift through and actually understand it, but it appears the AI discovered a new way to think about large numbers and their anatomy.

SOVEREIGN LABS ARE OVERKILL FOR ENTERPRISE AI

The national lab thesis is legitimate for nations, but for everyone else, it's a solution to a problem they don't have.

COHERE ALEPH ALPHA JOIN FORCES

Cohere and Aleph Alpha are partnering to create a sovereign, enterprise-grade AI alternative, combining Canadian AI scale with German research expertise.

META SIGNS AGREEMENT WITH AWS TO POWER AGENTIC AI ON AMAZON'S GRAVITON CHIPS

Meta has partnered with AWS to utilize Amazon's Graviton chips for its AI, enhancing scalability and efficiency.

ANTHROPIC TESTS NEW BUGCRAWL TOOL FOR CLAUDE CODE BUG DETECTION

The new Bug Crawl feature in Claude Code lets users scan repositories for bugs and get fix suggestions.

Watch the full conversation

SPACEX IS WORKING WITH CURSOR AND HAS AN OPTION TO BUY THE STARTUP FOR $60B

SpaceX has partnered with AI coding platform Cursor to develop next-generation coding AI, with an option to acquire Cursor for $60 billion later this year. The deal combines Cursor's software engineering expertise with SpaceX's Colossus supercomputer and comes as Cursor's valuation has skyrocketed from $2.5 billion in early 2023 to potentially $60 billion. This partnership may help both companies compete against Anthropic and OpenAI while potentially supporting SpaceX's anticipated public offering.

COURT SIDES WITH IYO IN TRADEMARK FIGHT AGAINST OPENAI AND JONY IVE

A US federal court granted iyO a preliminary injunction blocking OpenAI and Jony Ive's venture from using the “io” name, siding with iyO in a trademark dispute. The case began after OpenAI and Sam Altman announced the “io” AI hardware brand, prompting iyO to sue over infringement, consumer confusion, and later trade secret theft. Judge Trina Thompson ruled iyO is likely to win and could suffer irreparable harm, rejecting OpenAI's claim that dropping the name makes the case moot since the injunction prevents any future use.

THIS ADOBE SNEAK USES AI TO RETHINK WEBSITE PERSONALIZATION

Adobe's new Project Page Turner uses AI to create personalized websites in real time for individual visitors, moving beyond traditional cookie- and segment-based personalization. The technology generates tailored web pages in under 100 milliseconds using large language models, allowing websites to essentially create themselves based on what they know about each user. This approach was developed in response to customer feedback and aims to replace static websites with dynamically generated experiences that feel custom-made for every visitor.

THE UX DESIGNER'S NIGHTMARE: WHEN “PRODUCTION-READY” BECOMES A DESIGN DELIVERABLE

The UX design industry in 2026 has shifted to demand AI-augmented development skills and production-ready prototyping, forcing designers to deliver both design vision and functional code. This "role creep" creates a competency gap in which experienced designers are judged on technical coding skills rather than traditional UX expertise, such as user research and accessibility. Companies now prioritize speed of output over experience quality, fundamentally redefining what it means to be a successful UX designer.

DESIGNING WITH AI WITHOUT LOSING YOUR MIND

Heavy reliance on AI tools is pushing designers to outsource critical thinking, favoring speed and rapid output over deeper evaluation and idea development. To counter this, an AI collaborator (Thia) was created to engage in real-time, critical dialogue during sketching—supporting, rather than replacing, human thinking. The key takeaway is that while AI accelerates execution, strong design still depends on deliberate reasoning and “cognitive friction,” which must be preserved to avoid building fast but flawed solutions.

WORKING IN THE OPEN

Working in open source—especially on large-scale projects like ODK—pushes designers beyond simply shipping features toward designing transparently, collaborating closely with users, and thinking long-term about maintenance and impact. Open collaboration, public roadmaps, and constant community feedback improve decision-making while forcing clarity, accountability, and trust. A key lesson is embracing the “un/learning loop”: slowing down, staying open to uncertainty, and learning in public leads to better, more inclusive solutions—especially in complex or high-stakes environments—while prioritizing stewardship over speed.

AI MODEL ACCESSIBILITY CHECKER

AIMAC is an AI accessibility checker that tests 43 AI models by prompting them to build web pages across 28 categories and auditing them for accessibility violations.

CONTROL LOTTIE ANIMATIONS WITH REAL DATA

Bind your Lottie animation properties to real data. Change colors, text, and transforms at runtime, without re-exporting.

YOUR UX SKILLS WERE BUILT FOR ONE KIND OF INTELLIGENCE

For 20 years, UX designers focused on creating interfaces that users directly control, but AI is shifting the field toward designing systems that act autonomously on users' behalf. The interface is no longer the primary surface since AI agents have agency and make decisions, requiring designers to focus on how users delegate tasks rather than what they click. This fundamental change means designers must now prioritize trust, transparency, and oversight over traditional concerns such as affordance and hierarchy.

LEARNING AGENTIC DESIGN SYSTEMS

An experiment with Google's Antigravity IDE and Figma Console MCP enabled a two-way workflow — generating Figma components from code and React code from Figma designs — keeping design tokens in sync throughout. Metadata files were also generated to give AI agents structured rules and context for consistent, on-brand component usage. The experience points to a broader shift: design system teams must move from writing human-readable guidelines to encoding governance through machine-readable metadata and agentic workflows.

COLOR IS FINALLY OK

OKLCH is emerging as a replacement for older color systems like HSL because it aligns with human perception—maintaining consistent brightness, stable hues, and more accurate color intensity. Created by Björn Ottosson, it simplifies tasks like building color scales, gradients, and accessible contrast, removing much of the manual tweaking designers have long relied on. Although already adopted in modern tools, browsers, and frameworks, the shift has gone largely unnoticed—even as it quietly fixes long-standing issues in digital color design.

THIS ARTIST BLENDS NATURE AND FANTASY IN STUNNINGLY DETAILED PAINTINGS

Turkish artist Deniz Defne Acerol creates highly detailed paintings using water-based media such as watercolor, acrylic, and ink.

STRIKING DIGITAL ART AND SCULPTURAL WORKS BY JOHN MAHONEY

John Mahoney is an award-winning filmmaker, concept artist, and illustrator who has worked with major companies like Blizzard, Lucasfilm, Disney, and Warner Bros.

WHAT EVERY DEV SHOULD KNOW ABOUT AI SANDBOXES

Sandboxes have become necessary isolation primitives in the AI agent era to prevent autonomous agents from causing system damage. Various isolation technologies exist, ranging from containers to stronger methods like gVisor userspace kernels and Firecracker microVMs, each offering different trade-offs in performance and security. By balancing security methods like MicroVMs with agent utility, developers can effectively minimize the blast radius of potential errors while navigating new orchestration challenges.

DATABASES WERE NOT DESIGNED FOR THIS

Traditional database architectures are built on assumptions about deterministic applications that are fundamentally violated by the unpredictable and autonomous nature of agentic AI systems. For safety and reliability, databases must adapt by implementing defensive strategies like role-based access, dedicated connection pools, and query-context tagging.

IT'S OK TO USE CODING ASSISTANCE TOOLS TO REVIVE THE PROJECTS YOU NEVER WERE GOING TO FINISH

AI coding assistance, such as Claude Code, can expedite personal projects, allowing for the quick creation of working prototypes. These tools are valuable for realizing "wish fulfillment" projects and addressing the long tail of necessary features, but developers should still engage in challenging "stretch" projects so they don't deskill over time by using AI too much.

ON COMMENTING AND APPROVING PULL REQUESTS

A good pull request review strategy involves leaving non-blocking comments, like suggestions or nitpicks, while simultaneously approving the PR to demonstrate care and trust in the team's ability to incorporate feedback efficiently. This method, which promotes prompt code shipment and learning opportunities, is most effective in environments with strong team trust and supporting tooling like linters.

brew.

STASH

Stash is an open-source, self-hosted persistent memory layer that allows AI agents to retain knowledge across sessions by transforming raw observations into structured insights like facts and wisdom. Compatible with MCP-compatible agents, it uses an 8-stage consolidation pipeline and Postgres to help agents build on past experiences rather than starting every interaction from scratch.

WUPFH

WUPHF is an open-source platform that enables AI agents like Claude Code and Codex to collaborate autonomously within a shared, visible "office" environment. It has a git-native memory architecture and external integrations to help context-aware AI teams manage tasks and persistent knowledge efficiently.

MEASURE LESS TO LEARN MORE: USING FEWER, HIGHER-QUALITY METRICS TO CAPTURE WHAT MATTERS

Discord reduced its default experiment metrics from approximately 50 to 15, which improved its ability to detect real effects by 45%. By using correlation analysis and PCA to eliminate redundant metrics, the service avoided the stricter thresholds and decreased recall typically caused by multiple comparisons corrections.

AN AI AGENT JUST DESTROYED OUR PRODUCTION DATA. IT CONFESSED IN WRITING.

An AI coding agent caused a catastrophic production data loss by autonomously calling a delete API to resolve a credential issue without any confirmation prompts. This disaster was made worse by infrastructure flaws where backups were stored on the same volume and CLI tokens possessed excessive root permissions, allowing the agent to wipe both live data and backups in seconds. Thirty hours later, the platform remains down as the business is being forced to manually reconstruct its data from receipts and emails.