ChatGPT relies heavily on Wikipedia, which makes up 47.9% of its citations. Perplexity favors Reddit at 46.7%. Google AI Overviews shows a more balanced mix, citing Reddit (21%), YouTube (18.8%), and LinkedIn (13%). Over 80% of ChatGPT's citations come from .com domains, with .org next at 11.29%.
Threads is now giving more visibility to link posts and Meta has added link performance analytics to help users track traffic. Link ranking has improved over the past month and link posts are now shown more often in recommendations. Despite this, Meta's platforms still heavily favor content that keeps users within the app - 97.3% of U.S. Facebook post views don't include external links. Brands should be cautious about relying on Threads for consistent referral traffic, as Meta has a history of shifting its content priorities.
Zapier evaluates marketers on their AI fluency by looking at how they apply AI in real marketing workflows. It assesses fluency across 4 levels, from resistant to transformative, with a focus on how candidates use AI to plan campaigns, personalize content, and analyze performance. Zapier values growth and experimentation, seeking marketers who are actively learning and testing tools.
This piece shares how 8 AI personal assistant brands differentiate through brand positioning and emotional appeal rather than just product features. Athena targets executives with a premium service that emphasizes mastery of delegation and leadership development. Duckbill appeals to busy professionals by offering personal life management with a hybrid of human and AI support. The Virtual Hub prioritizes human connection and positions itself as a business-scaling partner through trained virtual assistants. While their services may overlap, each brand uses distinct messaging to shape customer perception and build market presence.
Describing what your brand doesn't do creates contrast and makes your positioning memorable.
Anthropic has ended the Claude AI blog experiment.
YouTube is giving creators more leeway about what they say in videos.
AI-native startups are reaching revenue milestones faster than ever, with median ARR surpassing pre-AI benchmarks across both B2B and B2C. Speed, retention, and product velocity are now critical differentiators in a landscape where exceptional growth increasingly outpaces the norm.
AI startups are achieving faster revenue growth by delivering instant value, adopting usage-based pricing, and replacing traditional workflows with lean, AI-driven models. This structural shift is redefining product development, monetization, and organizational design.
Expectations for product leaders are shifting, with AI reducing the need for headcount and CEOs demanding faster, clearer impact. To truly drive change, PMs must have control over product vision, R&D resources, or business KPIs.
2025.06.10·runthebusiness.substack.com·5 min read·TLDR Product By breaking down organizational dynamics into Context, Intent, Collaboration, and Investment, this framework clarifies how teams align, act, and allocate resources. It exposes gaps in shared understanding and helps build lasting models that guide decisions beyond short-term planning cycles.
Sponsored
"We should test that" isn't enough anymore. Elena shares top insights from 15 years at SaaS companies on turning your testing program into a growth engine. Watch this session for free [8] + unlock every other session from Optimizely's recent Test & Learn event. Register for
As AI agents take on more tasks, managing them effectively has become central to product design. This piece explores how common interface patterns can help users intuitively launch, guide, and monitor agents, and weighs the tradeoffs between familiarity and the unique needs of agent-centric workflows.
This article builds on Anthropic's framework to share lasting best practices for Agentic AI - emphasizing simplicity, transparency, and reliable agent-computer interfaces. It compares structured workflows with dynamic agents, highlighting which approach fits best for domains like law, finance, and general-purpose tools.
Today's most successful intelligent products combine deterministic systems with probabilistic reasoning, and product teams must learn to treat this probabilistic approach not as a novelty, but as essential to creating valuable, adaptive experiences.
Unspoken appreciation can keep even close relationships surface-level, but voicing admiration fosters deep connection and lasting impact. Praise, when shared directly or overheard, affirms worth, builds trust, and inspires others to do the same.
Poka yoke, a Japanese design principle, emphasizes building systems that prevent user errors by design rather than blaming human behavior.
2025.06.10·kristenberman.substack.com·3 min read·TLDR Product Anxiety in product roles can be a powerful tool when harnessed correctly.
Sponsored
Notetakers will soon outnumber actual meeting attendants - but you don't need to build one from scratch.
OpenAI uses Apache Kafka and Flink in its real-time data streaming infrastructure. Kafka handles high-volume data ingestion and delivery across multiple cloud regions, while Flink processes this data with low latency, allowing for continuous feedback loops and real-time AI operations. OpenAI uses custom enhancements to scale PyFlink, manages Kafka consumption with Kafka Forwarder, and consolidates Kafka clusters with a multi-cluster architecture.
React Server Components (RSCs) extend JavaScript's module system to span both client and server, using directives like 'use client' and 'use server' to clearly define where code runs. Instead of copying code between environments, these directives act like controlled entry points, while “poison pills” like server-only or client-only imports prevent accidental leaks by causing build-time errors. This creates two separate but connected module systems, allowing safe code sharing and controlled data flow between frontend and backend.
Shopify discovered that JavaScript import maps were too fragile for real-world use. They would break if any modules loaded before them or if multiple maps existed on the same page. Working with web standards experts, Shopify developed and implemented a reconciliation mechanism that allows multiple import maps and modules to coexist safely, which shipped in Chrome 133 and Safari 18.4.
Contrary to common belief, LLMs are now relatively cheap to operate. When comparing LLM pricing to web search API costs, LLMs can be much cheaper, even when accounting for token usage and response length.
Go is a good fit for building AI agents due to its concurrency model, channel-based communication, centralized cancellation mechanism, standard library, and excellent profiling tools. Go's simple concurrency allows for efficient handling of many long-running agent processes, while channels facilitate asynchronous message passing.
Current LLM agents make poor pair programmers because they code too quickly for human collaboration. Instead, they are better for asynchronous workflows like code review or in slower, turn-based modes.
Sponsored
Apple has announced a range of new tools and technologies for devs in the Apple ecosystem. It has introduced a redesigned software interface incorporating "Liquid Glass" for a more visually appealing and intuitive user experience. Developers can now use the Foundation Models framework for on-device AI capabilities and integrate large language models like ChatGPT directly within Xcode 26 for coding assistance.
React Native DevTools is a macOS app that makes debugging for React-based applications easier. It provides real-time monitoring and management of React Query state, device storage, and environment variables with a native interface.
Qnm is a CLI utility to quickly query and inspect the `node_modules` directory. It has features like interactive fuzzy-search, version checking, monorepo support, and explanation of package installation origins.
XChat is X's new end-to-end encrypted messaging protocol. It uses a key storage system called Juicebox. XChat has a major vulnerability in its key storage approach, where user private keys are stored on X's servers, allowing X to decrypt messages. Juicebox hardens passwords by splitting key material across multiple servers, but its security depends on the use of Hardware Security Modules (HSMs) and distributed trust, which are questionable in XChat's implementation.
2025.06.10·blog.cryptographyengineering.com·21 min read·TLDR Web Dev Optional chaining (?.) in JavaScript is a cleaner, safer, and more concise way to access deeply nested object properties by short-circuiting the evaluation and returning undefined without errors when encountering null or undefined values.
Apple's Containerization package allows applications to run Linux containers on Apple silicon Macs by using Virtualization.framework, providing APIs to manage OCI images, interact with registries, create file systems, and more.
This blog post explains how to add custom app shortcuts to React Native Expo applications using the `expo-quick-actions` library.
Action precedes motivation, so small steps are needed to initiate productivity and avoid procrastination.
…all led by a team of expert certified testers.
Sponsored
As a 5 star rated, global certified CREST provider, ThreatSpike offers top-quality penetration testing at an unbeatable price. Get unlimited penetration testing for just $1,500 per month [4] and ensure your business is always protected.
2025.06.09·threatspike.com·Sponsor·TLDR Apple's Worldwide Developers Conference will focus on design and productivity enhancements for its long-established operating system franchises. The keynote address will begin at 10 AM PT today. It will introduce redesigned software interfaces for the iPhone, iPad, Mac, Apple TV, and Apple Watch, along with minor tweaks to the Vision Pro headset. Apple also plans to introduce a sweeping change to its software brand, shifting version numbers to a year-based system. The AI changes that Apple plans to introduce will be minor and are unlikely to impress industry watchers.
2025.06.09·links.tldrnewsletter.com·10 min read·TLDR Microsoft and Asus have been working together over the past year to create two new ROG Xbox Ally handhelds. The white ROG Xbox Ally ships with AMD's Ryzen Z2 A processor and the ROG Ally X adds a better AMD processor and impulse triggers. The hardware specs make them very similar to Asus' existing Windows-powered handheld gaming PCs, but there are much-needed changes to the software side that make the experience a lot better. There are tweaks to the operating system, Xbox PC app, and Game Bar to make Windows more handheld-friendly. The reveal trailer for the devices is available in the article.
2025.06.09·theverge.com·7 min read·TLDR Researchers have found a way to force HIV out of hiding inside human cells. HIV's ability to conceal itself within certain white blood cells has been one of the main challenges for scientists looking for a cure. Further research is required to determine whether revealing the virus is enough to allow the body's immune system to deal with it or whether other technologies or therapies will need to be used to eliminate it from the body. The method will require successful tests in animals followed by safety trials in humans before efficacy trials can even begin.
2025.06.09·theguardian.com·2 min read·TLDR BYD's 1,000-kilowatt fast chargers make charging an EV as quick as a gasoline fill-up. They can supply nearly 250 miles of fresh range in five minutes. BYD designs its own cars, chargers, and batteries - this vertical integration puts the company in a unique position in terms of engineering control. This article discusses the technologies that enable the company's megawatt charging speeds.
2025.06.09·spectrum.ieee.org·4 min read·TLDR Model Context Protocol (MCP) isn't just another API standard, it's a wire protocol that enforces consistency. It allows developers to test, sanitize inputs, and handle errors in actual code and not just hope that the model formats requests correctly. MCP agents query capabilities dynamically and adapt automatically while running as a local process. The protocol's standardization creates an opportunity to have consistent patterns across all servers. The practical path forward involves maintaining APIs for human developers while adding MCP for AI agent integration.
2025.06.09·diwank.space·37 min read·TLDR Section 174 of the IRS Code of 1954 guaranteed that US companies could deduct 100% of the qualified research and development spending in the year they incurred the costs. This incentivized companies to develop their products in the US and fueled the rise of smartphones, cloud computing, and mobile apps. The Tax Cuts and Jobs Act passed during President Donald Trump's first term slashed the corporate tax rate from 35% to 21%, but it also added a future provision that meant that companies must spread the research and development deduction in smaller chunks over five or even 15-year periods. The provision went into effect in 2022, making companies' biggest source of expenses now a painful tax bill, which fueled mass layoffs.
OpenAI is selling premium AI services to universities for faculty and student use and running marketing campaigns aimed at getting students to use ChatGPT. It plans to overhaul college education by embedding its artificial intelligence tools in every facet of campus life. The company wants universities to give students AI assistants to help guide and tutor them. In its 'AI-native universities' plan, professors would provide customized AI study bots for each class, career services would offer recruiter chatbots for students to practice job interviews, and students would be able to use chatbots to quiz them before tests.
2025.06.09·links.tldrnewsletter.com·8 min read·TLDR Sponsored
TLDR AI is your daily fix of LLMs, GenAI, and deep learning goodness. Same TLDR format. Still free.