Search the archive

GET STARTED

STATEWRIGHT

Statewright uses state machine guardrails to regulate an AI agent's tool access during workflow phases like planning and testing, preventing errors and inefficient usage. This deterministic approach increases success rates on coding benchmarks by narrowing the problem space.

AGENTMEMORY

Agentmemory provides persistent, long-term memory for AI coding agents like Claude Code and Cursor, using a hybrid search system for high retrieval accuracy across development sessions. This centralized server architecture integrates with various agents, reducing token consumption by over 90%.

INSTRUCTURE PAYS RANSOM TO CANVAS HACKERS

Instructure paid an undisclosed ransom to ShinyHunters following two breaches of its Canvas learning management system, which compromised the personal data of approximately 275 million users and caused service disruptions. Though the company received a guarantee that the stolen data was destroyed, experts caution that paying the ransom incentivizes future attacks and offers no absolute certainty that the information will not be leaked.

BUILDING A NATURAL LANGUAGE INTERFACE TO THE SPOTIFY ADS API WITH CLAUDE CODE PLUGINS

The Spotify Ads API plugin for Claude Code allows users to manage complex Spotify advertising campaigns using simple natural language, which the tool translates into multi-step API sequences. This open-source system ensures transparent execution by using a Markdown-based architecture built on the raw OpenAPI spec.

installs. Try bitdrift.

SOFTWARE INTERNALS BOOK CLUB

The Software Internals Book Club is a global, email-based community where developers of all levels engage in expert-led discussions on high-caliber technical books covering databases, distributed systems, and software performance.

SHAPING THE FUTURE OF AI INTERACTION BY REIMAGINING THE MOUSE POINTER

A new AI-enhanced pointer reimagines human-computer interaction by using context-aware technology to allow users to navigate, edit, and command digital elements through intuitive gestures and natural language instead of complex text prompts.

LEARNING SOFTWARE ARCHITECTURE

Mastering software architecture depends on gaining practical experience and recognizing that organizational incentives and social dynamics shape a system's design far more than formal education or the code itself.

REPLACING A 3 GB SQLITE DATABASE WITH A 10 MB FST (FINITE STATE TRANSDUCER) BINARY

Transitioning a Finnish-English dictionary from a 3 GB SQLite database to a 10 MB finite state transducer reduced its storage requirements by 300x by using the data structure's ability to efficiently compress the frequent prefix and suffix patterns found in such languages.

See how it works →

WHO'S WINNING ENTERPRISE AI NOW: CLAUDE UP 128%, GEMINI UP 48%, OPENAI DOWN 8%, GROK STILL A ROUNDING ERROR

Enterprise adoption has shifted over the past twelve months. OpenAI is still in the lead, but its share has dropped to 56%. Claude adoption more than doubled in the last year. Gemini adoption grew from 27% to 40%, but Grok is barely growing within enterprise organizations. The coding assistant market is driving the numbers.

HOW AI CHANGES SOFTWARE P&L

Saas used to mean 70% gross margins. AI-native software might cap at 17%. Every AI call is personalized, so caching and multi-tenancy don't reduce your costs. Reasoning models burn 10-100x more tokens than the ones they replaced, so cost per task barely moves even as per-token prices drop. The math: 1 million users at $120 ARPU still leaves you at 17% gross, 11% net. That's consumer goods territory, not Salesforce. Four ways out: 1/10-cost clones (no moat), niche/lifestyle businesses (now viable since build costs collapsed), luxury software (Raya, Bloomberg Terminal), or vertical integration into atoms or services. If your deck still assumes 80% margins because that's what SaaS does, rewrite it this week.

HOW TO BECOME "AI-NATIVE"

AI-native is an organizational label, not a tech one. It is estimated that only roughly 1,000 companies above $5M ARR have actually rebuilt themselves so agents can operate inside the core workflows rather than at the edges. Most companies aren't legible to machines because pricing logic lives in "Final_v7_NEW" spreadsheets, refund policy hides in a Notion doc nobody trusts, and the sales process is "talk to Sarah," so the AI-native shift requires making implicit knowledge explicit: documented decision rules, agent-readable SOPs, clean customer objects, permissions, audit trails, and evaluation loops.

I CLOSED 70M IN ARR IN 2 YEARS. HERE'S EXACTLY HOW

Nicolas Brillante added over $70 million in ARR at Slash without a playbook. This post provides a bit of background on Brillante to explain who he is and how he achieved this feat. He started reselling digital items when he was 12 years old to buy a new computer. That early success kept him going, and he ended up managing multiple businesses.

HOW WINNERS PIVOT

A pivot is any change that rewires your unit economics, not a 180-degree rebuild. Before you change anything, map your entire conversion funnel in a FigJam, every marketing touchpoint, landing page, demo, lifecycle, trial, and retentive user, with the conversion percentage written between each step. The worst drop-off tells you which lever to pivot. Then test the pivot MVP-style with five different funnels for the five ICP groups you think will be stickiest, each with its own creative, landing pages, and email sequences. The worst mistake is broadcasting the pivot to the world before you have retentive paying beta testers.

agents, and their recommendations for startups. Watch the session

GITHUB COPILOT INDIVIDUAL PLANS: INTRODUCING FLEX ALLOTMENTS IN PRO AND PRO+, AND A NEW MAX PLAN

GitHub Copilot Pro and Pro+ plans now include more total usage at the same price through a new flex allotment. Paid plans now include base credits - once users go beyond that base, a flex allotment is applied automatically, and users can purchase more usage to keep going if they run out of tokens. Code completions and next edit suggestions remain unlimited on paid plans and don't consume credits. A new Max plan has been introduced for sustained, high-volume Copilot work.

AUTOMATE PROGRESSIVE ROLLOUTS WITH VERCEL FLAGS

Vercel Flags is a new tool for rolling out features to a growing percentage of users on a schedule. It supports progressive rollouts that gradually shift the traffic percentage to the new variant. Each stage has a target percentage and a duration. Exposing changes in stages allows developers to catch regressions on a small slice of users before they hit everyone.

PUBLIC STEALTH LEAVES OPPORTUNITY ON THE TABLE. WORKING IN SECRET IS EVEN WORSE

Founders often harm their prospects by staying too secretive about their efforts. Having no public trace of their startup work, such as a domain or LinkedIn update, can cause them to miss out on critical connections and opportunities. Basic visibility, like a personal domain or relevant LinkedIn job listing, helps signal commitment and legitimacy to potential investors and partners.

10X IS A LOT

There's a gap between claiming 10x and what 10x actually looks like in the wild. Real 10x is the gap where the old way stops making sense, email vs fax, Stripe vs wire transfers, not "23% faster but rounded up." Run the test before you write "10x" in your hero copy and ask whether your buyer would say the word unprompted on a call. If they wouldn't, you're padding.

case study.

AI CONTENDERS VS PRETENDERS IN ENTERPRISE

The enterprise AI buyer's filter has collapsed to one word: reliability.

MARC ANDREESSEN ON BUILDER CULTURE IN THE AGE OF AI (45 MINUTE LISTEN)

The doomer framing on jobs and productivity collapses once you observe what enterprise and consumer users actually do with the tools.

Download the free report →

WHO AMERICANS THINK IS COOL

56% of Americans say they are at least somewhat cool, including 16% who say very cool. Younger adults and Black Americans are more likely to feel this way, with 65% of those under 30 and 79% of Black Americans saying they are cool. Parents also rate themselves higher, with 64% of those with kids under 18 saying they are cool versus 54% of non-parents. Still, 74% say being cool is not important, and many think others care more than they do. Americans also rate themselves cooler now than in high school, at 56% vs 43%. When it comes to what is cool, most favor traditional interests like science at 87% and reading at 81%, while fewer see things like marijuana at 35% or sports betting at 25% as cool.

YOU'VE NEVER HAD YOUR BEST IDEA BY OPENING A NEW CHAT WINDOW

AI is weakening critical thinking, emotional depth, and originality when used without intention. Many brands adopt it reactively and prioritize speed over insight, leading to an overreliance on cognitive offloading, where people outsource the hard parts of thinking and communication. Research shows frequent AI use can reduce critical thinking, while creatives are pushed to produce more ideas faster instead of developing stronger ones. At the same time, globally trained models risk flattening culture by defaulting to familiar patterns, making human judgment and local perspective more important.

11 TIPS TO GET BLOGS CITED

Blogs that earn AI citations start with topics where AI answers are weak, tested using AI tools and search results before writing. Content is built around real user questions from sources like Google's People Also Ask feature and keyword research tools, then mapped into clear H2 and H3 structures. Each section opens with a direct answer followed by supporting detail for easy extraction. Posts include first hand examples, screenshots, and walkthroughs to demonstrate real experience. Writing stays simple and direct with clear visuals that explain key points. On page SEO includes optimized titles, headers, URLs, meta descriptions, and internal links. Technical setup includes proper indexing, schema, and regular audits. After publishing, content is updated, expanded with new questions, and tracked for AI citations and rankings.

5 THINGS TO DO BEFORE YOU DELETE YOUR FAQS

Google has removed FAQ rich results, but FAQ content still matters and shouldn't be deleted. Keep Q&A sections on pages to improve user experience and engagement, and recognize their growing value for AI tools like ChatGPT that pull from structured answers. Shift schema efforts toward formats Google still supports such as article or how-to markup, and turn high performing FAQ questions into standalone sections or pages to capture search intent. Rewrite FAQs to mirror how people phrase prompts in AI tools so they're more likely to surface in AI-driven search.

Plaid case study.

THE ORGANIC MEDIA MIX: THE STRATEGY DOCUMENT EVERY CMO SHOULD BE BUILDING RIGHT NOW

AI visibility now depends on a portfolio of organic channels, not just your website, so you need an Organic Media Mix that allocates resources based on where AI models actually cite brands. The framework starts with citation reports across tools like ChatGPT, Gemini, and Claude to identify which sources shape AI answers in your category, then scores those channels by influence, implementation difficulty, commercial intent, and sentiment. The data often overturns assumptions because citations can vary significantly by industry and brand. This framework enables teams to use citation data to guide ad placements, PR investment, influencer programs, and content production around the same high-intent queries.

SPRINGBOARDS

Springboards is a creative workspace built to generate a wide range of ideas across campaigns, not just a single output. It combines multiple AI models with features like customizable boards to spark different angles, a “voltage” control to adjust how safe or wild ideas get, pins to capture and remix ideas in real time, and an infinite canvas to organize and develop strategy collaboratively. It also includes reusable templates and flexible customization so teams can tailor workflows.

WE TRACKED 1,885 PAGES ADDING SCHEMA. AI CITATIONS BARELY MOVED

Despite the claim that JSON-LD schema is a direct AI visibility lever, an analysis of 1,885 pages found that it doesn't increase AI citations in Google AI Mode, ChatGPT, or Google AI Overviews. The study found a 2.4% lift in AI Mode and 2.2% lift in ChatGPT, but both results fell within statistical noise, while AI Overview citations dropped 4.6% relative to control pages. Schema correlates with stronger SEO operations rather than causing AI citation growth on its own. Sites that already invest in authority, technical SEO, and content quality still win AI visibility, even without schema changes. Schema still supports rich results and knowledge graph visibility, but brands should not expect it to drive measurable AI citation gains on already indexed pages.

TRADITIONAL MARKETING DOESN'T WORK ON AI SHOPPING AGENTS

A study simulating how AI shopping agents interact with product pages found that most e-commerce persuasion tactics refined for human cognition either fail or backfire when evaluated by AI. Of 8 promotional mechanisms tested across four models and over 16,000 shopping rounds, only ratings produced a consistently positive effect. Scarcity badges, bundling, and anchored pricing showed no reliable pattern, and advanced models like GPT-5 sometimes reacted negatively to aggressive promotional cues. The researchers recommend focusing on pricing and review fundamentals first, then building simulation environments to continuously test how each model responds.

BEHIND THE SCENES AT WIRECUTTER FOR AN EPIC DUEL OF AIR PURIFIERS

Wirecutter has maintained its lead in product recommendations by investing in original, hands-on testing where staff evaluate everything from air purifiers to mattresses. The publication has grown to 180 editorial employees and 15 million monthly readers since becoming part of a major news organization, more than tripling traffic in recent years. Reviewers operate behind a strict editorial wall with no visibility into affiliate revenue from their recommendations, a structure built on the principle that consumer trust is the only sustainable advantage. That independence has helped the brand weather an era in which competitors chasing affiliate revenue through SEO were penalized by algorithm updates targeting aggregated or low-expertise content.

BE TLDR'S FIRST VP MARKETING ($300K-$400K, FULLY REMOTE)

TLDR is hiring a VP of Marketing to scale growth, pipeline, and brand across 8M+ subscribers, 13 newsletters, and advertisers like AWS, Google Cloud, Microsoft, and Anthropic. Apply here [16].

YOUTUBE CREATOR PLAYBOOK - APRIL 2026

45% of YouTube Shorts users aren't on TikTok, and 65% aren't on Instagram Reels.

THIS A/B TEST RESULT WITH 99% CONFIDENCE IS 87% LIKELY TO BE WRONG

A landing page test claiming a 44% lift and 99% confidence likely produced a false positive because the experiment had just 6.9% statistical power instead of the recommended 80%.

TIKTOK TRENDS ARE BECOMING FORMATS, NOT JUST SOUNDS

TikTok trends are shifting from specific sounds to repeatable formats like "My Top 5 Horror", "God Forbid", or “Everything Hallelujah".

Schedule a demo to learn more

GOOGLE'S GEMINI OMNI VIDEO MODEL SURFACES AHEAD OF I/O DEBUT

Screenshots of a revised user interface showing a model card for Google's upcoming Gemini Omni video model were posted on Reddit over the weekend. Early user testing indicates the model has impressive prompt adherence, but video generation burned through credits quickly. Omni will likely be shipped in tiered variants (Flash and Pro).

MICROSOFT'S CEO INTERVENED WHEN OPENAI FIRED SAM ALTMAN, MUSK'S LAWYER CLAIMS

Elon Musk has accused Microsoft of aiding and abetting OpenAI as the lab abandoned its founding contract as a nonprofit. His lawyers have submitted evidence that Microsoft CEO Satya Nadella played a role in Sam Altman's efforts to return to OpenAI just five days after he was ousted as CEO. This was aimed at demonstrating Microsoft's significant control over OpenAI. Microsoft had earlier claimed that it should not have been drawn into the legal squabble between Musk and OpenAI.

FIGURE'S HUMANOID ROBOTS ORGANIZE ROOM, HANG CLOTHES, AND MAKE BED WITHOUT HUMANS

Figure has released a new video showing two robots performing household tasks in a coordinated bedroom-cleaning demonstration. The video (available in the article) highlights advances in humanoid collaboration, object handling, and domestic automation. Figure claims to have increased Figure O3 humanoid production from one robot daily to one hourly at its BotQ production facility in California.