Search the archive

ARCIMUS

Scan thousands of signals to surface the people who actually need you.

LIGHTBOX

Cursor for content creation.

SHELLCAST

Stream your vibe coding.

WHY RETENTION IS SO HARD FOR NEW TECH PRODUCTS

You might think fixing retention at a startup works like it does at big companies - run A/B tests, optimize onboarding, and send more notifications. However, that's attacking the wrong problem entirely. The truth is that bad retention means you built something people don't want, and no amount of optimization changes that. If your day-one retention is 10% instead of 40%, you need a complete pivot - redesign the home screen, change the core loop, and target different users. The only path to good retention is building something people already need to use frequently, then making it genuinely better than what exists.

USAGE BASED MODELS: WHY THEY SUCK AS A CUSTOMER

Software companies sold us on usage-based pricing with a simple promise - pay only for what you use. In practice, you buy a year's worth of cloud credits for $100,000, use only $50,000, then discover those unused credits disappear unless you spend another $100,000 next year. One founder called it "infrastructure hostage-taking" after going through exactly this. You ask to pay less since you're using less, they say no, and your unused credits become leverage for them to force a renewal.

AVEN RAISES $110M SERIES E TO EXPAND HOMEOWNER CREDIT PLATFORM

Aven secured a $110M Series E round at a $2.2B valuation to cut consumer credit costs, focusing initially on HELOC-backed credit cards. It has saved customers over $215M in interest and expanded credit lines to $3B. Aven plans to scale mortgage refinancing to provide a comprehensive financial platform for American homeowners.

YOU'RE DEFINITELY GOING TO BE A MANAGER NOW

AI is collapsing ten-person teams into two-person teams and killing middle management roles.

STRIPE CHALLENGER RAINFOREST LANDS $29M SERIES B

THE AI AGENT REALITY CHECK: WHY MANAGING ALWAYS-ON AI WORKERS IS YOUR NEXT BIG CHALLENGE

Managing 24/7 AI agents presents significant challenges despite their potential.

SEE STORAGE CONTROL IN ACTION

PLEX URGES USERS TO CHANGE PASSWORDS AFTER DATA BREACH

Plex disclosed a breach exposing user data, including usernames, emails, scrambled passwords, and some auth data. Though the passwords are unreadable, Plex urges users to reset and sign out of connected devices.

NO GAINS, JUST PAINS AS 1.6M FITNESS PHONE CALL RECORDINGS EXPOSED ONLINE

Security researcher Jeremiah Fowler uncovered an unprotected HelloGym database with 1.6 million audio recordings from major gym franchises, exposing customer and staff names, phone numbers, and sensitive conversations. Risks include voice cloning, deepfakes, and social engineering.

160,000 IMPACTED BY WAYNE MEMORIAL HOSPITAL DATA BREACH

Over 160,000 people had personal and medical data stolen in a ransomware attack on Wayne Memorial Hospital in May 2024. Hackers accessed sensitive personal, health, and financial information. The hospital responded quickly, restored systems, and offered affected individuals free credit monitoring.

LUNALOCK RANSOMWARE THREATENS VICTIMS BY FEEDING STOLEN DATA TO AI MODELS

LunaLock, a new ransomware group, has threatened to submit stolen artwork from the Artists&Clients website to AI companies for training datasets if its $50,000 ransom demand isn't met. This approach, a unique extortion technique, represents a dangerous precedent as it compromises victims' intellectual property permanently by incorporating stolen data into AI models, unlike traditional dark web leaks that may fade over time. The attack targets artists who are already vulnerable to AI data scraping by companies like OpenAI and Google.

GITHUB ACTIONS: A CLOUDY DAY FOR SECURITY - PART 1

Security challenges in GitHub Actions can affect even trusted teams, as attackers may exploit workflows or inject malicious code. Using branch protection, pull request approvals, environment variable management, and safeguards like the “four-eyes” principle reduces risks. Regular reviews and testing further strengthen CI/CD pipeline security.

MEMORY INTEGRITY ENFORCEMENT: A COMPLETE VISION FOR MEMORY SAFETY IN APPLE DEVICES

Apple's Memory Integrity Enforcement (MIE), introduced in the iPhone 17 and iPhone Air, combines secure memory allocators with Enhanced Memory Tagging Extension (EMTE) hardware to provide continuous memory safety against advanced spyware exploiting memory corruption. It uses hardware-enforced memory tagging to prevent buffer overflows and use-after-free attacks by assigning secret tags to memory and terminating processes when tags are incorrect. Tag Confidentiality Enforcement also defends against side-channel attacks, including Spectre V1 mitigations. Apple's tests show MIE defeats all recent real-world exploits, forcing attackers to restart development with new techniques and making memory corruption exploits more difficult and costly.

👋 Register now (San Francisco | Oct 8)

📘 Read Zenity Labs' AgentFlayer 0click Exploit research

VMDRAGONSLAYER

VMDragonSlayer is a framework for analyzing binaries protected by VM-based protectors like VMProtect 2.x/3.x, Themida, and custom malware VMs. It combines Dynamic Taint Tracking, Symbolic Execution, Pattern Classification, and Machine Learning to automate reverse engineering.

SOLA SECURITY (PRODUCT LAUNCH)

Sola Security has built an AI-powered no-code security platform that can design, customize, and deploy security tools in minutes.

DALEC

Dalec provides a declarative format for building system packages and containers from those packages. It is designed for building containers for Azure and supports Azure Linux 2 and 3 and Windows containers.

AI-POWERED CELEBRITY IMPERSONATION SCAMS: A THREAT INTELLIGENCE REPORT BY REDHUNT LABS

RedHunt Labs discovered a complex AI-driven investment scam targeting Indian users through Facebook and Instagram ads featuring deepfake videos of celebrities like Nirmala Sitharaman, Sadhguru, and Neha Kakkar promoting fake investment platforms. Victims lost an average of over ₹50,000. The scammers use deepfake technology to create convincing fake interviews and endorsements, bypassing Facebook's ad review system through short-lived campaigns and fake Amazon links. They redirect victims to counterfeit news websites impersonating major Indian media outlets like NDTV and India Today. The campaign involves a network of over 1,000 interconnected fraudulent websites and uses multiple payment methods, including UPI transfers and international bank accounts. It has expanded beyond social media to include compromised websites, fake Medium blog posts, and SEO manipulation to reach victims across multiple platforms.

NEPAL SOCIAL MEDIA BAN SPARKS PROTESTS, DOZENS INJURED

At least 14 people were killed and dozens injured in Nepal after police clashed with "Gen Z protesters" demonstrating against the government's ban on 26 major social media platforms, including Facebook, Instagram, WhatsApp, Signal, YouTube, and X. The restrictions were imposed after foreign tech companies failed to register locally under new regulations, citing concerns about cybercriminals spreading disinformation with fake accounts and unregistered platforms. Rights groups condemned both the ban and the police force, calling the restrictions a "dangerous precedent for press freedom" that could harm Nepal's business and tourism sectors, heavily reliant on social media.

TREASURY DEPARTMENT TARGETS SOUTHEAST ASIA SCAM HUBS WITH SANCTIONS

The U.S. Treasury Department sanctioned 19 individuals and organizations operating cyberscam hubs in Burma and Cambodia, responding to Americans losing at least $10 billion to Southeast Asian scam operations in 2024—a 66% increase from the previous year. The sanctions target notorious hubs like Shwe Kokko, Burma, where criminal organizations use debt bondage, violence, and threats of forced prostitution to coerce workers into conducting virtual currency investment scams and other online fraud against victims worldwide. The action aims to disrupt industrial-scale fraud operations that not only threaten Americans' financial security but also subject thousands of people to modern slavery conditions in forced labor compounds.

get the tool

MICROSOFT HITS PAUSE ON COPILOT ... IN SQL SERVER MANAGEMENT STUDIO

Microsoft temporarily removed Copilot functionality from SQL Server Management Studio (SSMS) 22 Preview 1 after 75% of survey respondents preferred GitHub Copilot integration over the current implementation.

VENEZUELA'S PRESIDENT MADURO SAID HIS HUAWEI MATE X6 CANNOT BE HACKED BY US CYBER SPIES

Venezuelan President Nicolás Maduro claimed his Huawei Mate X6 smartphone, reportedly gifted by China's Xi Jinping, is "the best phone in the world" and cannot be hacked by US cyber spies, despite cybersecurity experts noting that well-resourced nation-state actors can compromise any device and that HarmonyOS may contain undiscovered vulnerabilities due to limited global scrutiny.

KOSOVO HACKER PLEADS GUILTY TO RUNNING BLACKDB CYBERCRIME MARKETPLACE

Kosovo national Liridon Masurica, also known as "@blackdb," pleaded guilty to conspiracy to commit access device fraud for operating the BlackDB.cc cybercrime marketplace from 2018 to 2025.

FACEBOOK AD COSTS JUMP 21% IN 2025, BUT STILL BEAT GOOGLE

Facebook ads are getting more expensive for leads but cheaper for clicks. Cost per lead now averages $27.66, while CPC fell to $0.70. Traffic campaigns saw stronger engagement, making them a better ROI play in 2025. However, lead generation is under pressure with lower conversion rates. For marketers, this means tighter audience targeting, refreshed creative, and smarter funnel planning are essential for good ROI.

THE NEXT GENERATION OF CREATIVITY: A MAXIMALIST MOMENT RESHAPING CULTURE

Gen Z (14-24) is redefining media with Creative Maximalism, a participatory, visually complex, and internet-referential style shaped by platforms like YouTube, Minecraft, and Roblox. Teens spend 26% less time on TV and movies and 54% more time on social platforms and user-generated content than older audiences. 66% believe their generation has a big impact on online culture. They favor co-creation, immersive audiovisual storytelling, and content influenced by digital cultures.

5 SMART ECOMMERCE RETARGETING STRATEGIES TO RECOVER LOST SALES

Most e-commerce retargeting focuses too much on coupons, but buyers often abandon carts for reasons unrelated to price. Marketers should align messaging to each stage of the funnel. Cart abandoners may need social proof or urgency, while browsers need education or reminders. Use dynamic ads segmented by behavior and predicted value. A/B testing formats like SKU clusters versus single product reminders can significantly improve ROAS.

LINKEDIN ADS: MAX DELIVERY VS. MANUAL BIDDING

Test manual bidding on LinkedIn Ads instead of max delivery. In repeated campaigns, switching drove a 65.5% increase in clicks and 52.7% more impressions on the same budget, with only minor shifts in audience mix. Manual bidding helps stretch spend further and reach more new buyers, which is key for growth. However, keep in mind that chasing the lowest CPC/CPM can limit access to higher-value audiences.

TIKTOK SEARCH ADS CAMPAIGN: UNLOCKING INCREMENTAL GROWTH IN THE FUTURE OF SEARCH

TikTok's Search Ads Campaign delivers stronger results than traditional mid-funnel initiatives, driving 2.0x higher purchase lift overall and 2.2x for enterprise advertisers. In retail, dedicated search campaigns achieved a 1.9x purchase lift and higher incremental ROAS compared to non-search campaigns. TikTok searches are rising sharply, up 40% year over year, with 86% of Gen Z internet users searching on the platform weekly. Aerie demonstrated the impact of TikTok search ads, achieving a 3% lift in conversions, 3,800 incremental purchases, and a $2.64 iROAS by combining them with In-Feed Ads.

CMOS SEPARATING THE HYPE FROM WHAT'S REAL (WEBINAR)

Exit Five founder Dave Gerhardt will host a live webinar on September 23 with Jennifer Delevante-Moulen (CMO of Knak) and Tara Robertson (CMO of Bitly) to cut through AI hype and share how B2B marketing teams are actually using AI today. They will discuss where AI accelerates execution, which creative work still needs humans, and how CMOs can update playbooks for AI.

WHATAGRAPH 3.0

Whatagraph replaces fragmented tools with one system. Marketers can connect 55+ sources, clean and organize data, and create branded reports in seconds using prompts or drag-and-drop. Its AI assistant delivers instant insights, summaries in 18 languages, and plain-language answers to ad-hoc questions. Reports auto-update every 30 minutes.

THE FUNNEL HASN'T COLLAPSED, IT'S COMPRESSED

Social commerce has compressed into a single scroll. Platforms like TikTok and Amazon collapsed discovery, intent, and purchase into one swipe. This isn't the death of the funnel but its evolution, where mental states still matter, only processed faster and often simultaneously. The risk is that brands overinvest in visible conversions while neglecting memory-building, even though long-term growth still depends on penetration and brand availability. The real challenge is balancing speed with stickiness so brands aren't forgotten when discovery and purchase happen in seconds.

HOW CONTENT LABELS INCREASE CLICKS

Researchers found that content labels such as “most read” and “most shared” significantly affect click behavior. Across 5 experiments, 2 surveys, and 36,679 ad clicks, people perceived “most read” content as 18.4% more informative, while “most shared” was seen as 32.5% more entertaining. When searching for information, people clicked 23.9% more on articles labeled “most read” and were 67.9% more willing to read them compared to “most shared.” For entertaining content, “most shared” labels drove higher clicks, showing that labeling content with the right signal can increase engagement.

CHATGPT PROMPTS TO REMOVE CRM DUPES

Templated prompts to scrub duplicates in your data.

1 IN 4 DOLLARS IS MADE IN PEAK SEASON, OR NOVEMBER AND DECEMBER

Despite talk of earlier holiday shopping, most ecommerce dollars still arrive in November and December.

TOP EMAIL DESIGNS FOR AUGUST

Design inspo for welcome emails, renewal reminders, and more.

→ READ THE POST

SPEEDING UP FIREFOX LOCAL AI RUNTIME

Mozilla has accelerated Firefox's AI Runtime by replacing the WebAssembly-based “onnxruntime-web” with native C++ ONNX Runtime integrated directly into Firefox. This change delivers 2-10x faster inference performance for AI features like PDF alt-text generation (improving from 3.5 seconds to 350ms) and smart tab grouping, while removing WebAssembly warm-up overhead.

SHOULD YOU PRELOAD FONTS FOR PERFORMANCE?

Font preloading can improve website performance by eliminating double paints and layout shifts. However, it can also delay First Contentful Paint (FCP) since Chrome now blocks rendering until preloaded fonts arrive (with timeouts of 100-1,500ms). It's best to preload only essential above-the-fold fonts (maximum two), self-host them, and prioritize text fonts over icon fonts for better user experience.

PARTYTOWN: OPTIMIZE THIRD PARTY SCRIPTS WITH WEB WORKERS

Partytown is a library that optimizes website performance by automatically moving third-party scripts (like analytics, ads, and chat widgets) into web workers, allowing them to run in background threads instead of blocking the main thread responsible for rendering and user interactions. This approach can improve Core Web Vitals scores. One example showed a Next.js site's Lighthouse performance score jumping from 70 to 99 after implementing Partytown with Google Tag Manager.