Search the archive

👉 EXPLORE THE AGENDA HERE

SPOILERJS

spoilerjs is a lightweight, dependency-free web component for creating spoiler text with a Telegram-inspired particle effect. It allows users to hide text behind a spoiler warning, revealing it with a configurable particle animation upon clicking.

QQQA

qqqa is a stateless CLI tool providing LLM assistance in the command line through two binaries: `qq` for quick questions and `qa` for single-step agent tasks. It supports OpenAI and Groq APIs. The tool has simple formatting, configurable profiles, and safety rails for file access and command execution.

DEXTO

Dexto is a toolkit for building AI agents that turn natural language into real-world actions by orchestrating LLMs, tools, and data into persistent systems. It has a configuration-driven framework with session management, tool orchestration, and multimodal interactions. The toolkit supports over 50 LLMs, various storage options, and provides built-in observability.

THE PULSE: AMAZON LAYOFFS – AI OR ECONOMY TO BLAME?

Amazon announced another round of mass layoffs affecting 14,000 employees, with software engineers being disproportionately impacted, following previous cuts of over 27,000 jobs since 2023. While Amazon claims the layoffs are to become "more lean" and invest in AI transformation, analysis shows that the company has strong financials with $93B in cash reserves and $32B in free cash flow, making the $2-4B in salary savings from layoffs relatively small.

1,500+ PRS LATER: SPOTIFY'S JOURNEY WITH OUR BACKGROUND CODING AGENT (PART 1)

Spotify has successfully integrated AI coding agents into its Fleet Management system, which has generated over 1,500 merged pull requests for automated code maintenance tasks like dependency updates, language modernization, and complex migrations. The AI-powered approach has reduced manual coding time by 60-90% compared to traditional methods and allows engineers to define fleet-wide changes using natural language prompts rather than complex programmatic transformations.

MANY POSTS HERE ARE TEACHING BEGINNERS WRONG LESSONS ON SMALL DATA

Some Kaggle competitions often teach beginners to overengineer solutions, leading to overfitting and worse leaderboard scores than simpler models due to the small dataset size.

OPENPCC

OpenPCC is an open-source framework for provably private AI inference that lets users run AI models without exposing sensitive data by using encrypted streaming, hardware attestation, and unlinkable requests.

SANDDANCE

SandDance is an open-source data visualization tool that uses unit visualizations and animated transitions to help users explore, understand, and present data.

HOMARR

Homarr is a customizable, open-source dashboard application with numerous integrations, user management, and other features for self-hosted applications.

growth. Read more.

VENTURE CAPITAL IS NOT COMPETITIVE

Venture capital no longer behaves like a competitive market. The language of rivalry persists, but the substance has thinned. The same firms crowd around the same deals, rehearsing the same convictions until they harden into consensus. Somewhere along the way, the infinite game became finite. What began as a search for outliers has become an exercise in pattern recognition. Capital pools, performance converges, and the illusion of competition masks a quiet retreat from risk. The question is not who wins this round, but whether there's still a game being played at all.

THE $100B QUESTION: HOW SAAS GIANTS ARE REWRITING THE RULES OF VALUE WITH AI IN 2025

SaaS pricing ran on predictability for two decades. Seats and flat plans made sense when value scaled with people. AI broke that logic - one agent can do the work of ten, and each query can carry a volatile compute bill. Companies are now rebuilding their models in real time. Hybrid pricing buys time but not clarity. Outcome pricing promises fairness but demands proof. Between the two lies a growing realization that software was predictable only because the people using it were.

SAM ALTMAN ON OPENAI'S REVENUE

OpenAI's annual revenue run rate has surpassed $20B. Future infrastructure commitments are reaching $1.4 trillion. The company's plans include enterprise offerings, AI cloud services, and ventures into consumer devices and scientific discovery.

HOW YOU.COM SURVIVED CHATGPT WITH A PIVOT TO ENTERPRISE AI

You.com shifted from consumer search to enterprise AI, offering search APIs and complex data analytics to businesses following ChatGPT's launch. Competing with Google and others in the AI market, You.com leverages its own search infrastructure to deliver detailed reports and agent-based solutions like business analytics and financial analysis. The company also introduced Auto Mode for integrating multiple LLMs and secured $100M in VC funding to expand its agent-centric services.

WHY OPENAI WANTS A PIECE OF AMD

OpenAI will buy up to 6 GW of AMD GPUs, starting with 1 GW of MI450 in H2 2026, and get $0.01 warrants for up to 160M shares that vest with 1–6 GW deployments and other milestones, potentially reaching ~10% ownership. It chose AMD because Nvidia supply is tight and pricey, AMD claims MI450 will beat current Nvidia parts, and AMD's smaller base trades equity for multi‑GW orders. Monetization trails GPU demand, especially in video, and tariff‑driven inflation cuts rate‑cut odds, raising downside risk for AI‑levered names.

LOOK FOR YES

Product and legal teams often clash due to misunderstood risk levels. Risk managers should ensure their advice isn't seen as an absolute veto but rather as guidance by framing context and trade-offs. Product managers need to actively seek solutions to legal concerns, aiming for responsible risk-taking rather than defaulting to "no."

👉 Register now to watch it live.

VENTURE CODEX

An investor database for startups looking to raise.

SIGMA MIND

Build human-like conversations with AI

AGNES AI

Agnes AI is a collaborative AI agent that lets teams co-edit, share memory, and build slides, reports, and strategy docs together in real time.

KOAH LAUNCHES: ADSENSE FOR GENAI

Koah introduces AdSense-like monetization for GenAI apps by integrating context-aware ads into chat. With proven success in apps like Liner and Luzia, developers see high revenue without compromising user experience, averaging $10 ECPM and 3-5x higher engagement rates.

THE LOST ART OF CRYING

Former Facebook VP Julie Zhuo just wrote the most vulnerable thing you'll read about founder psychology this year. She reveals how Silicon Valley's "ice queen" culture (the obsession with logic, optimization, and emotional control) is actually a fear response, not a superpower. As AI takes over rational work like calculations and planning, the only irreplaceable human skill is our ability to truly feel: grief, awe, connection, vulnerability. Zhuo discovered this through what she calls "Big Cries" using music and solitude, and argues that in an era where we're building superhuman machines, learning to cry again isn't weakness but reclaiming the last thing AI can never touch.

THE INTERNET WANTS TO BE FRAGMENTED

The social media empire is crumbling, and Gen Z is leading the exodus. Nearly half of Gen Z plans to spend less time on social networks in 2025, abandoning the "town square" platforms for intimate Discord channels, DMs, and curated feeds. The big experiment of throwing humanity into one digital room failed. This shift back to small, self-selected communities is a massive opportunity for builders creating tools for micro-communities instead of trying to be the next Twitter.

SUBTLE COMPUTING'S VOICE ISOLATION MODELS HELP COMPUTERS UNDERSTAND YOU IN NOISY ENVIRONMENTS

Subtle Computing's voice-isolation models enhance voice recognition in noisy environments, benefiting AI products and services.

HOW TO GET FEEDBACK

When was the last time someone told you the truth about your work?

STOP HIRING PEOPLE

Founders love to hire - it feels like momentum, but most hiring starts from impulse instead of from need.

has had on security and efficiency.

Register now

HYUNDAI AUTOEVER AMERICA DATA BREACH EXPOSES SSNS, DRIVERS LICENSES

Hyundai AutoEver America, which provides IT services for Hyundai and Kia affiliates, reported a data breach. The breached data includes names, SSNs, and driver's licenses. The disclosure did not state whether the breach only included employee data or also included customer data.

MALICIOUS ANDROID APPS ON GOOGLE PLAY DOWNLOADED 42M TIMES

Researchers from Zscaler report that an estimated 239 malicious apps were downloaded 42 million times from the Google Play Store between June 2024 and May 2025. Zscaler noted a 67% year-over-year growth in malware targeting mobile devices during this period. Zscaler has identified three malware families that had a notable impact on Android users: the Asta banking trojan, the Android Void Android TV backdoor, and the Xnotice RAT.

400,000 WORDPRESS SITES AFFECTED BY ACCOUNT TAKEOVER VULNERABILITY IN POST SMTP WORDPRESS PLUGIN

Over 400,000 WordPress sites are at risk due to a severe vulnerability in the Post SMTP plugin, which allows hackers to view password reset emails and potentially take over accounts. Exploitation is already underway. Users must urgently update to version 3.6.1 to prevent their site from being compromised.

CREDENTIALS AND MISCONFIGURATIONS BEHIND MOST CLOUD BREACHES, SAYS AWS

The 'Building Cloud Trust" report by AWS and Vanson Bourne surveyed 2,800 organizations across 13 countries, revealing that about 80% experienced data breaches in the past year, with common attack vectors including vulnerability exploitation, compromised credentials, misconfigurations, and physical theft. Attackers are increasingly using stolen credentials through methods like VPN/VDI abuse, AI-enabled phishing, and legitimate tool exploitation, with a predicted shift toward targeting non-human identities protected by single-factor authentication. To combat these threats, security teams should focus on cloud security posture management, adopt phishing-resistant MFA, enforce least privilege access, and prioritize patching internet-facing infrastructure, while acknowledging human error as a primary cause.

BREAKING INTO GITLAB: ATTACKING AND DEFENDING SELF-HOSTED CI/CD ENVIRONMENTS

Attackers can abuse globally scoped GitLab instance runners to execute arbitrary code and pivot into cloud environments if runners are configured with shell executors and attached IAM roles. Once a runner is compromised, the attackers can scrape build artifacts and environment files to steal secrets, SSH keys, or cloud credentials. Strategic defense centers on eliminating global runners, enforcing containerized job isolation, tightening IAM role scoping, and restricting access to the GitLab instance to reduce blast radius.

WINDOWS INTERNALS: SECURE CALLS - THE BRIDGE BETWEEN THE NT KERNEL AND SECURE KERNEL

Secure Calls are a mechanism in Windows that allows the “regular” (NT) kernel, residing in virtual trust level (VTL) 0, to communicate with the secure kernel (SK) operating in VTL 1. Secure calls are implemented as a hypercall with the hypervisor brokering the connection. In this post, the author reverse-engineers the secure call process and structure and introduces a tool for making secure calls with user-supplied arguments.

CURL-IMPERSONATE

curl-impersonate is a special build of curl that can impersonate Chrome, Edge, Safari, and Firefox.

UNDELETE

This tool recovers deleted NPM packages by querying five registries (npmjs.org, cnpmjs.org, npmmirror.com, huaweicloud.com, and tencent.com) to download 1-20 recent versions with metadata, including usernames, emails, and maintainer information, utilizing auto-retries and security filtering. Security researchers can retrieve discontinued malicious packages for malware analysis and threat intelligence. Features include custom output, silent mode with JSON, and no external dependencies, making deployment easy. Teams use this for post-incident recovery, building malware collections, and investigating suspicious packages that have been quickly removed from NPM.

MALANTA (PRODUCT LAUNCH)

Malanta is a cybersecurity startup that utilizes AI to detect digital traces left by attackers during their attack setup, forecasting, and dismantling malicious infrastructure before attacks occur. Its technology provides warning in advance and proactive defense for clients.

HOW THE F5 BREACH, CISA JOB CUTS, AND A GOVERNMENT SHUTDOWN ARE ERODING US CYBER READINESS

A China-linked breach of F5's BIG-IP software exposed source code and undisclosed vulnerabilities, creating a path for custom exploits against critical infrastructure and government networks. At the same time, significant workforce cuts at CISA and a federal shutdown are degrading incident response capacity, election protection efforts, and national coordination. Security teams should assume reduced federal support and prioritize proactive hardening, continuous monitoring, and supply chain risk management across high-impact edge and network devices.

CYBERSOCEVAL: BENCHMARKING LLMS CAPABILITIES FOR MALWARE ANALYSIS AND THREAT INTELLIGENCE REASONING

Researchers from Meta and CrowdStrike introduced the open-source CyberSOCEval benchmark to assess large language models (LLMs) on two critical SOC tasks: malware analysis and threat intelligence reasoning. Technical findings show that current models score only ~15-28% on malware analysis and 43-53% on threat-intelligence reasoning, highlighting significant gaps in interpreting JSON logs, mapping attack chains to MITRE ATT&CK, and reasoning about multi-hop adversaries. Security teams should be cautious about off-the-shelf LLM reliance, instead using these benchmarks to validate models, augment human analysts rather than replace them, and prioritize training with domain-specific cyber data.

META IS EARNING A FORTUNE ON A DELUGE OF FRAUDULENT ADS

Internal documents from late last year predicted that Meta would earn about 10% of its overall annual revenue ($16 billion) from fraudulent ads. Meta's automated system only bans advertisers if it is 95% certain that the ads are malicious. Otherwise, the advertisers are simply charged a higher rate. Documents also show that users who click on these scams or malicious ads are more likely to see them in the future.

COURT REIMPOSES ORIGINAL SENTENCE FOR CAPITAL ONE HACKER

A federal judge reaffirmed a non-custodial sentence for the engineer behind the massive Capital One breach, maintaining supervised release and restitution rather than prison time despite appeals arguing the punishment was too lenient.

M&S PEGS CYBERATTACK CLEANUP COSTS AT £136M AS PROFITS SLUMP

Marks & Spencer faced a £136 million cleanup bill for a cyberattack, which hurt profits, as the retailer struggled with halted online operations and expensive recovery efforts, resulting in a drop of over 55%.

SONICWALL FIREWALL BACKUPS STOLEN BY NATION-STATE ACTOR

Nation-state attackers breached SonicWall's cloud backup, stealing firewall configs but no other data.

👉 Try the AI CMO for free →