Search the archive

Extract Structured Data

ANTHROPIC'S MODEL NAMING, EXTRAPOLATED

Anthropic's literary-inspired model naming convention is funnily projected into a future where AI products scale from simple verses to increasingly complex and absurd narrative formats.

IF CLAUDE FABLE STOPS HELPING YOU, YOU'LL NEVER KNOW

Anthropic's implementation of invisible performance restrictions on AI development tasks creates a supply chain risk because developers can no longer distinguish between a model's natural failure and intentional, silent policy enforcement.

PIONEERING THE AGENTIC SHIFT WITHIN SALESFORCE ENGINEERING

Salesforce transitioned to a fully agentic software development lifecycle by standardizing on Claude Code and removing token limits.

AI CREATIVITY AND DISCOVERY

True scientific discovery and creativity in AI require a transition from just mimicry of generative models toward systems that use variation, evaluation, and selective retention to produce results that are both novel and effective.

Get your copy

AMAZON'S DATA LOOP JUST GOT STRONGER

Amazon has expanded its CTV advantage by combining commerce-based targeting with premium streaming inventory and new B2B audience data. Advertisers can now use Amazon shopping and intent signals to target Netflix viewers, which turns CTV into a more measurable channel for brands that sell on Amazon by linking impressions to product views, add-to-carts, and purchases. Amazon also integrated LinkedIn's 1+ billion member dataset into its DSP, which gives B2B marketers access to role, seniority, and industry targeting on streaming TV at scale. The move strengthens both audience precision and attribution at a time when streaming accounts for 38% of TV viewing in the UK and US programmatic CTV spending has reached $37 billion.

INSTACART FOUND HONESTY ABOUT STOCKOUTS LIFTS SPENDING 5.8%

A field experiment involving roughly 1.2 million Instacart shoppers found that proactively warning customers about low-stock items increased spending 5.8% over six months, driven primarily by higher order frequency rather than larger basket sizes. Warned shoppers were 25% less likely to select the flagged item, switching to better-stocked alternatives instead, which reduced out-of-stock refunds by 2.6%. The findings reframe stockout warnings from a conversion risk into a trust signal, with customers returning more often to platforms they perceive as reliable.

80% OF YOUR B2B DEMO PIPELINE IS ALREADY ON YOUTUBE. HERE'S WHAT TO DO ABOUT IT

YouTube has become a high-intent B2B acquisition channel, not just an awareness platform, and many SaaS brands overlook organic video opportunities that compete directly with expensive paid search. Analysis of 8,566 SaaS keywords found that YouTube appears on page one for 80.2% of "demo" searches, along with roughly 20% of "review," "vs," and "best" queries where buyers actively evaluate vendors. The strategy is to prioritize YouTube content for demos, comparisons, reviews, trials, and educational searches, then align video topics with the same high-value keywords targeted in SEO and Google Ads. Brands can reduce dependence on CPCs that often range from $20 to $49 by creating videos that rank organically and continue generating traffic over time. The biggest opportunity lies in specific, long-tail content because YouTube gains visibility again on longer searches that mirror the detailed queries increasingly generated by AI search tools.

BRAND PILLAR TEMPLATE: TRIED & TRUE VS. TRYING NEW

Jack Appleby's dual-track content pillar system separates proven ideas from experimental ones. Structure 60-80% of content around “tried and true” pillars based on past performance and established industry patterns. Reserve 20-40% for “trying new” pillars that are still on brand but more exploratory. This framing reduces perceived risk by guaranteeing stable performance from core content while justifying creative testing. Over time, evaluate results quarterly and promote any successful experiments into the core pillars, then refresh the experimental set to keep testing new directions.

LAST MINUTE PRIME DAY PREP

Prime Day on June 23 is a narrower affiliate opportunity than it once was, with AI Overviews absorbing commercial search intent. Publishers already in the space should prioritize hub pages with clean internal linking, granular keyword monitoring as interest typically peaks around 10 days before the event, and live updates that keep content fresh during the sale itself. For those without an established affiliate vertical, building one around a single tentpole event is not a viable strategy.

[13]and all things AI in our AI hub.

HOW TO CREATE A PREFERRED SOURCES LINK

Google's Preferred Sources feature lets users increase the likelihood of a website appearing in AI Overviews and AI Mode by marking it as a preferred source. This can also earn a “Preferred” label in results. To set this up, first create a Google preferences link using this format: “google.com/preferences/source?q=yourwebsite.com”. Then distribute that link across newsletters, social posts, and your website. When users click it, they check a box to select your site as a preferred source, signaling to Google to surface your content more often.

OPTIMIZELY TEST + LEARN (VIRTUAL EVENT)

This free 2-hour event on June 17 will explore how AI is reshaping experimentation for product and growth teams. It will feature live demos of AI agents that support ideation, test building, prioritization, and results analysis. Speakers will also share practical frameworks teams can apply immediately. The keynote is by Elena Verna, who shares insights on the gap between AI hype and real experimentation practice and what teams need to adapt.

WHEN AI-WRITTEN MESSAGES BACKFIRE

People react negatively when they know a brand used AI to write emotional marketing messages, especially announcements, apologies, and founder-style notes. Across six experiments with more than 2,200 participants, AI-written emotional messages reduced brand recommendations by up to 24.6% and increased feelings of moral disgust by 58.4%. The backlash is strongest when companies present AI-generated copy as if it came from a real person. Avoid using AI for emotionally charged communications, and instead use it only as a light editing tool while keeping the final voice human, specific, and authentic.

FIVE FRIENDS MAKING COMEDY TOGETHER PROVE A CREATOR BRAND DOESN'T NEED A SINGLE STAR

What Happened Last Night is a five-person comedy collective that has grown to over 100K Instagram followers without a single front-facing creator. The team relied on one highly produced video per week made by longtime friends. They spent months building a backlog before launch, then settled into a tight cycle of writing and filming that prioritizes cinematic storytelling over volume.

THE BRAND ARCHITECTURE BEHIND AUDEMARS PIGUET X SWATCH

The Audemars Piguet x Swatch Royal Pop Collection is a co-branding exercise in managed desirability that pairs Audemars Piguet's heritage and scarcity with Swatch's accessibility and cultural irreverence. The collection created a controlled entry point into the luxury brand without touching the core product. For younger audiences who cannot yet access a Royal Oak, the collaboration offers cultural proximity to the brand at a price point that preserves rather than undermines the aspirational distance.

HOW SLATE'S HEAD OF MARKETING EDITS HER SOCIAL VIDEOS (3 MINUTE VIDEO)

A walkthrough of how to edit a “moving postcard” using Blackmagic Camera and Slate.

GOOGLE IS TESTING SPONSORED SHOPS IN SERPS

The new Shopping ad format groups multiple products from a single retailer into one branded unit alongside store ratings and brand signals, shifting competition from the product level to the retailer level.

NEW YORK NOW REQUIRES DISCLOSURE OF AI PERFORMERS IN ADS

New York became the first state to legally require disclosure of AI-generated performers in advertising.

Read the auth.md docs

INITIAL IMPRESSIONS OF CLAUDE FABLE 5

Anthropic claims that Claude Fable 5 offers the same performance as Mythos 5 but with stricter guardrails. It has a 1 million token context window with 128,000 maximum output tokens. The model is priced at twice the price of the Claude Opus models at $10 per million input tokens and $50 per million output tokens. The Claude API has new mechanisms for alerting users when these guardrails are activated, and there is an option to request to fall back to another model automatically if something gets rejected.

ELON MUSK'S FIRST-GEN ORBITAL DATA CENTER CRAFT SPANS WIDER THAN A BOEING 747 AND RUNS AN INTERCHANGEABLE CHIP PAYLOAD

SpaceX's AI1 satellite will have a 70-meter deployed wingspan and carry a 150 kW peak compute payload. It will have an interchangeable hardware design that will let different chipmakers supply the processors. The craft will operate at roughly 600 km. SpaceX has already filed with the FCC to launch up to a million orbital data center satellites.

NASA'S NEXT MOON MISSION IS A RUBE GOLDBERG MACHINE OF CORPORATE FAILURE POINTS

NASA has announced the crew for its upcoming Artemis 3 mission. The mission will require the astronauts to travel to Earth's orbit inside NASA's Orion capsule, where they'll meet and board both Blue Origin's Blue Moon lander and SpaceX's Starship in quick succession over three days. It is designed to lay the groundwork for the crewed Moon landing tentatively scheduled for 2028. NASA is confident that Blue Origin's New Glenn will be ready for the mission despite the recent explosion that dealt significant damage to the company's launch pad.

WORLD'S FIRST WIND-POWERED UNDERWATER DATACENTRE STARTS OPERATING IN CHINA

The Shanghai Lingang undersea datacenter demonstration project is a joint effort between HiCloud Technology and China Communications Construction, a state-owned company. The datacenter, which has a capacity of 24 megawatts, is submerged 10 meters below the surface of the water. It is powered by a nearby offshore windfarm. The datacenter uses less energy compared to land-based datacenters because of the natural cooling effect that comes from being submerged in seawater.

collabs. See the future of influencer campaigns

CONTAINER

container is a tool for creating and running lightweight virtual Linux containers on Mac. It consumes and produces OCI-compatible container images and can pull and run images from any standard container registry. The tool uses the Containerization Swift package for low-level container, image, and process management.

KTX

ktx is an executable context layer for data and analytics agents. It allows AI agents to query data accurately and with full context. ktx is self-improving and can teach agents how to query data warehouses accurately. It learns from company knowledge, maps the data stack, builds a semantic layer, and serves agents at execution. A video explaining the tool is available in the repository.

WHAT IT FEELS LIKE TO WORK WITH MYTHOS

Anthropic's Mythos-class AI model, Fable, represents a very real leap over past models. In informal experiments, it outperforms every other public model by a considerable margin. The model has guardrails that prevent it from being used for things like cybersecurity, but it can still produce impressive results from single prompts. This post looks at several examples of outputs from the model and describes what it is like working with it.

APPLE WINS CONSUMER AI BY DEFAULT

Apple is set to win AI from a consumer perspective. The company may seem late to the AI race, but it was just doing what it has always done: watch new products and services develop and then jump in later with a better user experience. Its latest product announcements may seem underwhelming to the people living at the edge of AI, but most people don't need or even want that. Apple is set to bring AI usage mainstream with Apple Intelligence and the new Siri AI - if it is able to roll out what it showcased at WWDC.

GM EYES NEW BATTERY CHEMISTRY TO GROW AI DATA CENTER, ENERGY STORAGE BUSINESS

General Motors plans to increase its vehicle-to-grid capabilities and develop next-generation sodium-ion batteries, which have the potential to operate without active cooling and with much less system complexity.

THE IPHONE'S LAST STAND

While other companies are building infrastructure for AI, Apple is making the iPhone the true core of Siri, allowing it to not spend billions in capex.

TEST-CASE REDUCERS ARE UNDERAPPRECIATED DEBUGGING TOOLS

Test-case reducers are tools that automate the process of reducing test cases, which often makes debugging vastly easier.

GRAB YOUR SPOT.

DATATYPE

Datatype is a variable font that turns text into charts.

WWDC26 — THE SMALL THINGS

A list of the smaller changes announced at WWDC.

WHY DEVELOPERS USE LLMS TO WRITE BLOG POSTS

A lot of people who use LLMs to write have never written before, and while most perform substantial editing, they don't feel that LLMs capture their voice or ideas.

Save your spot to see how

CLAUDE FABLE 5 LAUNCH

Anthropic announced Claude Fable 5 for general use and Claude Mythos 5 for selected cyberdefenders and infrastructure providers. The models were described as highly capable across software engineering, research, vision, and cybersecurity, with conservative safeguards applied to Fable 5.

FLUID, NATURAL VOICE TRANSLATION WITH GEMINI 3.5 LIVE TRANSLATE

Gemini 3.5 Live Translate is an audio model for real-time speech-to-speech translation across 70+ languages that eliminates awkward pauses and maintains natural intonation. It is rolling out via Google products, including Meet for private preview and Google Translate on Android and iOS, enhancing multilingual communication.

GOOGLE'S BACKSTOPS UNDERPIN $35 BILLION CHIP DEAL FOR ANTHROPIC

Google is supporting Anthropic's $35 billion chip lease by backstopping payments at five data centers. This financial backing highlights intricate business alliances between major tech companies in the AI sector. Anthropic's role in this significant financing was previously undisclosed.

TEXT AS A SERIOUS OPTIMIZATION LAYER

Prompts, context, memory, retrieval stores, and harnesses function as real update mechanisms. The piece frames text optimization as sample-efficient learning and a new axis for update-time compute.

IMPLICATIONS OF LARGE-SCALE TEST-TIME COMPUTE

Benchmark grids hide the real story because LLM capability is now a function of test-time compute, illustrated by GPT-5.5 looking only marginally better than GPT-5.4 on max-compute cyber evals but substantially stronger once tokens, cost, or latency are controlled on the x-axis. The performance plateau is now empirically very far out and stronger models push the plateau further, so single-scalar benchmark scores will only get less informative each release.