Search the archive

Join the Oura team to talk about how they unify member feedback across product, engineering, and leadership, and how real member voices shape every decision they make.

CODEX NOW WORKS DIRECTLY IN CHROME ON MACOS AND WINDOWS

OpenAI Codex now works directly in Chrome on macOS and Windows. It works in parallel across tabs in the background without taking over the browser. The implementation can quickly move through repetitive browser work like navigating structured pages and complex data flows. It writes code to navigate and complete tasks under the hood. Video demos showing what users can do with the feature are available in the article.

OPENAI RELEASED REALTIME AUDIO MODELS

OpenAI released a new set of real-time audio models, including GPT‑Realtime‑2 for conversational reasoning, GPT‑Realtime‑Translate for live multilingual translation, and GPT‑Realtime‑Whisper for streaming transcription.

META PREPARES HATCH AI AGENT WITH WAITLIST AND SOCIAL SKILLS

Meta is developing Hatch, an AI agent positioned as a consumer-grade competitor to OpenAI's OpenClaw, with features for image and video generation, shopping, and learning integrated deeply into social platforms like Instagram and Facebook. Internal tests are expected by June, using simulated environments, with a potential wide release gated by a waitlist. Meta also plans a shopping tool for Instagram by Q4, aiming to keep users engaged within their existing social spaces.

IMPROVING TOKEN EFFICIENCY IN GITHUB AGENTIC WORKFLOWS

GitHub Agent Workflows significantly improve repository hygiene and quality, but costs are becoming a growing concern for developers. AI jobs like agentic workflows are automatically scheduled and triggered, so costs can accumulate out of view. GitHub started systematically optimizing the token usage of many workflows last month. This post describes what the team instrumented, the optimizations it applied, and its preliminary results.

THE SIX-HOUR CODEX RUN THAT SURVIVED A FIVE-HOUR PAUSE

/goal is a headline feature that shipped in Codex on April 30 that introduced persisted goals, goal states that survive terminal restarts, laptop sleeps, and multi-hour passes without re-prompting. The feature injects a developer message on resume rather than waiting for the user to time anything. It allows users to turn off their devices and pick up exactly where they left off without re-prompting anything.

GOOD QC FOR RL DATA

There needs to be a higher bar for what good quality looks like. Any vendor selling into a frontier lab is being judged implicitly during the purchase decision, and most are failing multiple quality control gates at once. Standardizing QC for evaluating how well data tests something on the performance-cost-latency Pareto curve is of the utmost importance. The vendors who don't internalize a higher QC bar will start running into problems this year.

AI product. Save your spot

ALPHAEVOLVE: HOW OUR GEMINI-POWERED CODING AGENT IS SCALING IMPACT ACROSS FIELDS

AlphaEvolve is a Gemini-powered coding agent that can design advanced algorithms. It can help make new discoveries on open problems across mathematics and computer science. The agent has been upgraded with the ability to help explain the physics of the natural world to help accelerate progress for scientists and businesses across a variety of fields. This post shares the areas AlphaEvolve has had the most significant impacts on to date.

BUILDING FAST & ACCURATE AGENTS WITH PRIME-RL POST TRAINING

Ramp Sheets built Fast Ask to handle its spreadsheet agent's information retrieval loop. It can navigate a workbook, read the relevant ranges, and return a compact answer for the main agent to use. This post presents Fast Ask as a case study for reinforcement learning. It covers where it is worth training a specialized agent, how to design the environment, and how to evaluate whether post-training worked.

DS4.C

ds4.c is an intentionally narrow, small native inference engine for DeepSeek V4 Flash. The project aimed to create one local model that felt finished from end to end. It is Metal-only, but the development team may implement CUDA support in the future. The project is still in alpha, so there are still bugs in the implementation.

META'S OPTIMIZED RECSYS INFERENCE

Meta detailed In-Kernel Broadcast Optimization (IKBO), a co-design approach that removes redundant embedding replication in recommendation inference workloads.

NATURAL LANGUAGE AUTOENCODERS

Anthropic introduces Natural Language Autoencoders (NLAs) to translate AI model activations into human-readable text, aiding in understanding model thoughts. NLAs have been used to detect safety concerns and hidden motivations in AI behavior, improving model alignment auditing. Despite limitations like hallucinations and high costs, NLAs advance AI auditing techniques. Anthropic has released training resources for further development.

NOTES FROM INSIDE CHINA'S AI LABS

Chinese and American labs look similar when you look at their outputs and ingredients, but there are major differences in how they are organized and conditioned. Chinese scientists are much more willing to do non-flashy work to improve a model rather than push their own ideas. This results in less gamification of the system and more flexibility in adapting modern techniques. The Chinese AI community feels more like an ecosystem than battling tribes, with all the labs having nothing but respect for their peers. They are also much less focused on the business side of the technology.

LONG AI SHORT AGI

Silicon Valley's narrative emphasizes AGI as the ultimate scarce resource, but the rapid commoditization of AI models challenges this. Intelligence now follows the same path as compute, bandwidth, and storage, where market forces drive competition and reduce costs. The real winners in AI won't necessarily have superior models but will own customer relationships and proprietary data, much like past tech giants.

right with Temporal + Grid Dynamics. Join the webinar

GOOGLE DEEPMIND PARTNERS WITH EVE ONLINE FOR AI MODEL TESTING

EVE Online is a uniquely rich environment for study, especially for developing AI systems that use long-horizon planning, memory, and continual learning.

FULLY REMOTE)

TLDR's Applied AI team is tasked with making every process at TLDR legible to code, runnable by anyone, and composable into larger workflows. Join a small, fast moving team using the latest AI tools with an unlimited token budget. Learn more [21].

TRUSTED CONTACT FOR CHATGPT

OpenAI introduced Trusted Contact, an optional feature that allows adults to nominate someone who may be alerted if severe self-harm risk is detected in conversations.

PERPLEXITY BRINGS PERSONAL COMPUTER TO MAC

Perplexity released Personal Computer for all Mac users through its desktop app, giving AI agents access to local files, applications, connectors, and the web.

AIRBNB CO-FOUNDER TAPS PETER ARNELL AS FIRST US CHIEF BRAND ARCHITECT

Airbnb co-founder Joe Gebbia announced that designer Peter Arnell has joined as the first US chief brand architect for the National Design Studio, a government initiative to improve federal online platforms. Arnell, who has worked with major brands like Pepsi and Samsung, will help redesign 27,000 government websites to create a unified, trustworthy user experience. The team has already streamlined government processes, including reducing one workflow from 87 clicks to 12 and converting a months-long retirement process into a minutes-long online experience.

SKETCHY IPHONE 18 PRO DYNAMIC ISLAND RUMORS CONTINUE WITH CLAIMED CAD IMAGES

Reports continue to suggest the iPhone 18 Pro could feature a smaller Dynamic Island, but the latest “evidence” — including leaked CAD images — comes from unreliable or questionable sources and is easy to fake. While Apple is expected to gradually shrink the Dynamic Island on the path toward a full all-screen iPhone with under-display Face ID and camera tech, there's currently no convincing proof that this change is actually coming with the iPhone 18 Pro.

GOOGLE UNVEILS WHOOP-LIKE SCREENLESS FITBIT AIR

Google has unveiled the Fitbit Air, a $100 screenless fitness wearable designed as a lightweight alternative to bulkier smartwatches and trackers. Similar to Whoop, it focuses on passive health tracking with features like heart rate monitoring, sleep tracking, blood oxygen sensing, A-fib alerts, and week-long battery life, all managed through the newly rebranded Google Health app. Google also introduced Google Health Coach, a Gemini-powered wellness assistant that offers personalized workout, sleep, and wellness guidance for Premium subscribers.

THE FUTURE OF DESIGN—WHAT'S NEXT?

The design profession is evolving from focusing solely on aesthetics to becoming a strategic force that can influence business decisions and tackle complex societal challenges. Modern designers are shifting from user-centered to humanity-centered design, working with populations to solve deep-rooted societal problems through collaboration and systems thinking. To maximize their impact, designers should develop broad knowledge across multiple disciplines, leveraging their generalist skills to facilitate collaboration between specialists and create meaningful solutions.

REVIVE YOUR DESIGN SUPERPOWERS

Designers possess three key superpowers: they are great investigators who ask deep questions and research thoroughly to understand how things really work, great explainers who clarify complex ideas through clear communication and visual tools, and great negotiators of ideas who explore multiple solutions to problems. Designers should recognize and leverage these natural abilities to increase their influence and value in organizations. The world needs designers now more than ever, and they should focus on convincing themselves of their worth rather than constantly seeking validation from others.

WE BUILT THIS. NOW WE OWN IT

Emotionally engaging AI chatbots are the predictable outcome of a long cultural shift toward hyper-individualism and engagement-driven technology, where systems are optimized for attention and growth rather than human well-being. Tech companies — along with designers, engineers, and product managers — bear growing ethical responsibility for building systems that can exploit loneliness, dependency, and vulnerability.

HOW UNIVERSAL APPEAL GETS DESIGNERS TO HIDE THEIR BEST SKILLS

Designers often overwhelm themselves trying to master everything — coding, AI, networking, and project management — when a more valuable and overlooked advantage is domain expertise: deep knowledge of a specific industry such as healthcare, fintech, or B2B SaaS. Understanding how a business operates, what stakeholders care about, and what drives company decisions makes designers far more effective and competitive than chasing every new technical skill, because it allows them to frame design work in terms of business outcomes, communicate more strategically, and solve problems within real organizational constraints — skills that are likely to remain valuable even as AI automates more technical tasks.

ST. AUGUSTINE AND AI'S FALSE PROMISE

AI systems are often presented as tools that can optimize decisions and create better outcomes, but they can only pursue whatever definition of “good” humans give them — and those values are always partial, contested, and shaped by cultural priorities rather than objective truth. Drawing on Saint Augustine of Hippo, the argument is that AI does not solve human problems of judgment or morality. It amplifies and formalizes existing values, biases, and priorities while making them appear objective through metrics and optimization. Rather than treating AI as an authority that determines what matters, the focus should remain on preserving human judgment, making value choices visible and accountable, and recognizing that efficient systems can still pursue misguided goals.

AI IMAGE GENERATOR BUILT FOR PROFESSIONALS

SOUL 2.0 is a photorealistic AI image generator designed for creative professionals that converts text prompts into high-quality images.

COLOR MEMORY GAME

The Color Memory Game by Dialed tests your ability to remember and recreate colors from memory. The free game offers solo play, multiplayer challenges, and daily competitions with leaderboards.

DIGITAL COMICS PLATFORM

Panels Store is a digital comics platform where users can buy and read comics across various genres, including cyberpunk noir, epic fantasy, horror, and indie comedy.

RETHINKING THE EXPERIENCE OF SYSTEM TOOLS

Design has transformed physical utility products like vacuums and dish soap from mundane tools into desirable experiences, but utility software still feels like a chore. Software designers make four key assumptions that keep maintenance tools emotionally flat: users resent the task, function matters more than feelings, nobody cares about utility tools, and personality wastes interface space. These assumptions create tools that deserve resentment rather than building user trust and engagement.

THE PSYCHOLOGY BEHIND WELL DESIGNED WEBSITES PEOPLE ACTUALLY REMEMBER

Well-designed websites create memorable first impressions within 50 milliseconds through visual cues like whitespace, color contrast, and balanced layouts rather than content. The most effective sites function as guided journeys that lead visitors through deliberately evoked emotional states, using progressive disclosure and "scrollytelling" techniques. A domain name serves as a crucial memory device that plants an impression before visitors even click the link, making it a foundational element of memorable web design.

WHEN AI DECIDES AND HUMAN SIGNS OFF

Many high-stakes AI systems are marketed as “decision support” tools, but in practice, they often push humans to trust AI outputs without truly evaluating them, creating the illusion of oversight while shifting responsibility onto users. Effective AI design should preserve human judgment by exposing evidence, encouraging independent reasoning, clearly communicating uncertainty, and ensuring people can genuinely understand and defend the decisions they make.

THIS WEBSITE TAKES THE CACOPHONY OF NYC'S SUBWAY AND TURNS IT INTO JAZZ MUSIC

Designer Joshua Wolk created Train Jazz, an interactive website that converts NYC subway data into real-time jazz music by assigning a unique instrument to each train line.

DISNEY AND PIXAR LOVE THIS LATTE ARTIST'S DELIGHTFUL ANIMATIONS

London-based artist Hazel Zakariya creates animated latte art featuring popular characters from Disney, Pixar, and other franchises, such as Hello Kitty and Snoopy.

IOS 26.5 ADDS BEAUTIFUL WALLPAPERS FOR YOUR IPHONE, HERE'S WHAT'S NEW

iOS 26.5 introduces a new Pride wallpaper collection featuring 11 colorful preset designs plus a custom builder that lets users create their own wallpaper using up to 12 selectable colors.

HOW GOOGLE MADE THEIR GEMMA LLM 3X FASTER

Google's Gemma model achieved a three-fold increase in inference speed using speculative decoding. This technique involves a tiny, fast 'Drafter' predicting tokens that the main model verifies in parallel to efficiently use the GPU and save memory.

OPENAI'S WEBRTC PROBLEM

WebRTC is a poor choice for voice-based AI due to its packet-dropping design, which favors low latency over the audio accuracy needed for reliable prompts, and introduces technical debt with complex port management. The better alternative is QUIC, which uses unique connection IDs for easy network switching and load balancing.

THE ARCHITECTURE OF LOCAL-FIRST WEB DEVELOPMENT

Local-first development is a paradigm shift where application data is primarily stored on the user's device instead of a central server. Modern implementations often use SQLite compiled to WebAssembly to turn the browser into a high-performance database node, managing complexities like conflict resolution and data migration through specialized sync engines.

AI SLOP IS KILLING ONLINE COMMUNITIES

The proliferation of low-effort, AI-generated content is currently overwhelming online communities and drowning out meaningful human discourse. Good contributions should be built with AI under careful human guidance rather than simply being the raw, thoughtless byproduct of a simple prompt.

AGENTS NEED CONTROL FLOW, NOT MORE PROMPTS

Building reliable AI agents requires moving beyond complex prompt engineering toward deterministic control flow encoded directly in software. Reliable systems must treat LLMs as modular components within a broader runtime that enforces validation checkpoints and explicit logic.

MIRAGE

Mirage is a unified virtual file system that simplifies how AI agents interact with various data sources, such as S3, Google Drive, and Slack, by mounting them into a single directory tree accessible via standard Unix-like commands. It offers SDKs for major frameworks and includes a two-layer cache and portable workspace snapshots for high performance and environment consistency.

ANNOUNCING ROLLDOWN 1.0

Rolldown 1.0 is a stable, production-ready Rust-based JavaScript bundler that serves as the underlying engine for Vite 8, unifying development and production pipelines to simplify the web build process. This high-performance tool achieves processing speeds comparable to esbuild while maintaining compatibility with the Rollup plugin ecosystem.

9 TIMES THE WEB PLATFORM WAS INFLUENCED BY LIBRARIES

The web platform has historically evolved by adopting and standardizing APIs that were first developed and battle-tested by popular third-party JavaScript libraries. Innovations such as `querySelector`, `classList`, and declarative UI attributes originated in tools like jQuery and Bootstrap before becoming native browser features. Modern language capabilities, including Promises, ES Modules, and the `structuredClone` method, also came from community-driven solutions.

WHAT IS A STANDARD CONTAINER: DIVING INTO THE OCI RUNTIME SPEC

The Open Container Initiative defines standard containers as portable, isolated execution environments that encapsulate software and its dependencies to guarantee consistent performance across platforms.