Search the archive

CURSOR SUPPORT RUNS ON CURSOR

Cursor's support team now investigates most tickets directly inside Cursor by combining code search, logs, documentation, and internal tools through MCP integrations.

CONSTITUTIONAL BLACK-BOX MONITORING FOR SCHEMING IN LLM AGENTS

Black-box monitors can detect AI scheming using only observable actions, without internal insights.

LLMS CAN UNMASK PSEUDONYMOUS USERS AT SCALE WITH SURPRISING ACCURACY

Recent advances in AI capabilities have made it clear that there is an urgent need to rethink various aspects of computer privacy.

OPENAI CODEX REVIEW 2026 — UPDATED FROM DAILY USE

Codex in 2026 has become production-ready infrastructure and is ready for serious development work.

I'VE DECIDED TO LEAVE OPENAI

Max Schwarzer, VP of Research and Head of Post-Training at OpenAI, has left OpenAI to join Anthropic.

Make your first 90 days count. Download the playbook.

AI CODING STARTUP CURSOR HITS $2 BILLION ANNUAL SALES RATE

Cursor is one of the fastest-growing startups of all time. It is less than five years old, yet its product is already ingrained in many software programmers' daily routine. The company's annualized revenue topped $2 billion in February. About 60% of the revenue is coming from corporate customers. The startup was valued at $29.3 billion in November, making it one of the most valuable AI startups in the US.

ANTHROPIC VS. WHITE HOUSE PUTS $60 BILLION AT RISK

The $60 billion investment into Anthropic from over 200 venture capital investors is now at risk due to a contract dispute with the Pentagon. Anthropic being designated a supply chain risk is unprecedented. It will prevent other military contractors from deploying Claude in their applications. It could also require companies like Nvidia, which do business with the US military, to sever their commercial activities with Anthropic. The situation is becoming an existential moment for all of American AI and its investors.

ANTHROPIC CLAUDE EXPERIENCED WIDESPREAD OUTAGE

Anthropic faced broad service disruptions affecting Claude.ai and Claude Code, with many users unable to log in, while the company reported that the Claude API continued operating normally.

CHINESE AI FIRM MINIMAX MORE THAN DOUBLES REVENUE IN FIRST POST-IPO RESULTS

Chinese AI startup MiniMax's annual revenue soared to $79 million in 2025 from $30.5 million the previous year. Net loss widened to $1.87 billion from $465.2 million. Analysts were expecting the equivalent of about $72 million in revenue and a net loss of roughly $564 million. These are the first results published since the company went public in January. Its shares have increased more than fourfold from their IPO price, pushing the company's market capitalization past $30 billion.

KEEPING COMMUNITY HUMAN WHILE SCALING WITH AGENTS

THE STATE OF CONSUMER AI. PART 1 - USAGE

AI apps have now crossed 1 billion weekly active users, with ChatGPT accounting for 900 million. ChatGPT is the only AI app with both the installed base and new user momentum to become the core utility. Its lead is even more pronounced across DAU:MAU ratios, retention, and time spent. The challenge now is to see whether the usage is deepening into genuine habit or still behaving like a utility people visit briefly and leave.

ALIBABA'S SMALL, OPEN SOURCE QWEN3.5-9B BEATS OPENAI'S GPT-OSS-120B AND CAN RUN ON STANDARD LAPTOPS

Alibaba recently unveiled its Qwen3.5 Small Model Series. Qwen3.5-0.8B and 2B are intended for prototyping and deployment on edge devices where battery life is paramount. Qwen3.5-4B is a strong multimodal base for lightweight agents with a 262,144 token context window. Qwen3.5-9B is a compact reasoning model that outperforms OpenAI's open source gpt-oss-120B on key third-party benchmarks. The weights for the models are available now under Apache 2.0 licenses on Hugging Face and ModelScope.

DIFFUSION LANGUAGE MODEL

dLLM consolidates training and evaluation workflows for diffusion-based language models into a single library, aiming to improve transparency and reproducibility across the development pipeline.

JUST-IN-TIME AGENTIC MEMORY FRAMEWORK

GAM is a new memory system that generates optimized runtime contexts for LLM agents using a two-part design: a lightweight Memorizer and a dynamic Researcher.

THE OTHER SIDE OF ADS IN CHATGPT: ADVERTISER PERSPECTIVE

This article provides details on how OpenAI's ads program works for advertisers. Some of the companies invited to trial ads in ChatGPT include Adobe, Target, Albertson, and WPP Media. Each was asked to commit to at least $200,000. The companies selected specific words that have to be prompted by users for the business to be advertised. Advertisers are managed manually.

ANTHROPIC CEO RESPONDS TO TRUMP ORDER, PENTAGON CLASH (27 MINUTE VIDEO)

Anthropic CEO Dario Amodei sat down with CBS News for an exclusive interview hours after Defense Secretary Pete Hegseth declared the company a supply chain risk to national security, which restricts military contractors from doing business with the AI giant. Amodei called the move "retaliatory and punitive," and he said Anthropic sought to draw "red lines" in the government's use of its technology because "we believe that crossing those lines is contrary to American values, and we wanted to stand up for American values."

Miro webinar. Secure your spot

LLMFIT

llmfit is a terminal tool that right-sizes large language models to fit a system's RAM, CPU, and GPU.

GOOGLE TESTS PROJECTS FEATURE FOR GEMINI ENTERPRISE

Google is testing a project feature for Gemini Enterprise that allows users to organize chats by topic and set goals for AI interactions.

REINFORCEMENT LEARNING FOR CUDA KERNEL OPTIMIZATION

CUDA Agent introduced a large-scale agentic reinforcement learning system that synthesizes data, augments execution with specialized skills, and trains over long horizons to generate high-performance CUDA kernels.

Get the whitepaper →

OPENAI AGREEMENT WITH THE DEPARTMENT OF WAR

OpenAI announced a classified deployment agreement with the US Department of War that included explicit red lines against mass surveillance, autonomous weapons control, and high-stakes automated decision-making.

OPENAI RAISES $110B

OpenAI announced a $110 billion funding round at a $730 billion pre-money valuation backed by Amazon, Nvidia, and SoftBank to expand compute, distribution, and enterprise infrastructure. The company now supports 900 million weekly active users, 50 million consumer subscribers, 9 million paying business users, and 1.6 million weekly Codex developers, extending AI use deep into workflows across business functions. This capital and strategic cloud partnerships aim to shift frontier AI from research to global production scale with broader enterprise and developer adoption.

GOOGLE TESTS NEW LEARNING HUB POWERED BY GOAL-BASED ACTIONS

Google accidentally revealed a new Gemini feature, "Goal Scheduled Actions," that allows AI to autonomously adjust tasks toward defined objectives. This contrasts with existing scheduled actions that repeat fixed prompts. The feature aligns with Google's LearnLM initiative, suggesting applications in education, where structured AI guidance could aid skill development.

"ALL LAWFUL USE": MUCH MORE THAN YOU WANTED TO KNOW

Anthropic was recently declared a supply chain risk due to its refusal to allow the Department of War to use its technology for mass surveillance and autonomous weapons. A few hours later, the Department of War announced a partnership with OpenAI. While OpenAI and the Department of War have tried to assure the public that they won't use AI for mass surveillance and autonomous weaponry, there are wide loopholes in current laws against their use. The Department of War is also able to change many of the rules that currently exist at any time.

WHY XML TAGS ARE SO FUNDAMENTAL TO CLAUDE

Structuring prompts with XML can be a transformative experience in Claude. The AI's framework specifically incorporates XML tags as key elements. The repurposing of XML technology may be a core aspect of what makes Claude distinctive. It gives Claude the ability to distinguish between the transition from first-order to second-order expressions, which is a mechanism fundamentally required for information transfer between any two entities. Claude's awareness of the concept of delimiters is crucial to every processing and communication of information, and it is this capacity that makes Claude so effective at interpreting layered meaning.

90% OF EXPERT WORK CAN'T BE VERIFIED BY TODAY'S AI TRAINING METHODS

Around 90% of expert work across healthcare, legal, finance, and engineering relies on subjective judgment, making it incompatible with current RLVR-style verification. To force verifiability, teams over-specify tasks and rubrics, turning real expert reasoning into shallow instruction-following and corrupting the training signal. Verification, not data scale, is the core bottleneck, and the winners will be those who can evaluate judgment-heavy, non-deterministic work without distorting it.

CLAUDE FOR CHROME EXTENSION INTERNALS (V1.0.56)

when they sign up. Get your free year

THE THIRD ERA OF AI SOFTWARE DEVELOPMENT

Cursor described a shift toward autonomous coding agents that operate over longer time horizons with minimal supervision, marking a “third era” of AI-assisted development. The company reported that over one-third of its merged pull requests were generated by cloud-based agents and outlined a future where developers manage fleets of agents rather than write code directly.

MCP IS DEAD. LONG LIVE THE CLI

MCP is already dying. Major projects like OpenClaw and Pi don't support it for good reason. LLMs are really good at figuring out things on their own. All they really need is a command-line interface and some docs. CLIs are much more practical for both humans and agents. The tools already exist and are well-documented, and both humans and agents understand how to use them.

NEW AGENT-BROWSER SKILL: ELECTRON

agent-browser can now control desktop apps built with Electron. It can also be used to debug Electron apps. agent-browser can be added to any coding agent. Popular Electron apps include Discord, Figma, Notion, Spotify, and VS Code.

ANDREW NG SAYS AGI IS DECADES AWAY—AND THE REAL AI BUBBLE RISK IS IN THE TRAINING LAYER

Andrew Ng, founder of DeepLearning.AI and Coursera, executive chairman of Landing AI, and founding lead of the Google Brain team, says that AI capable of performing the full breadth of human intellectual tasks remains decades away. He recently appeared in an interview where he discussed enterprise adoption of agentic AI, whether AI is in a bubble, the AI infrastructure build-outs, geopolitical fragmentation and its effects on global AI strategy, and more. This post contains a transcript of that interview.

WHEN AI LABS BECOME DEFENSE CONTRACTORS

Government contacts offer predictable, multi-year, politically protected revenue streams that don't churn when a competitor releases a better model. Any lab serious about classified work has to build an organizational structure that meets government operational security requirements, and the lab that builds that structure first has a moat that no competitor can cross quickly. The defense budget is the largest single purchaser of advanced technology on the planet, and it's growing. Its multi-year contract cycles reward incumbents, and it's willing to use blunt regulatory tools against companies that don't cooperate.

PERPLEXITY INTEGRATED AT SYSTEM LEVEL ON SAMSUNG GALAXY S26

Perplexity was integrated directly into Samsung's Galaxy S26 at the OS level, powering both its own assistant and Samsung's Bixby.

AGENTS WILL PAY LIKE LOCALS, NOT TOURISTS

Agents will pay for services using pre-negotiated B2B terms and credit, bypassing traditional retail payment rails.

AI WON'T AUTOMATICALLY ACCELERATE CLINICAL TRIALS

Drug development is bottlenecked by issues that cannot be solved by technology alone.

IS AI DOING LESS & LESS?

AI's role in workflows has shifted from full autonomy to specific tasks, with 65% of nodes now running as deterministic code.

Download the ebook

NANO BANANA 2

Google's Nano Banana 2 (Gemini 3.1 Flash Image) combines the advanced reasoning and image quality of Nano Banana Pro with significantly faster generation speeds.

STATEMENT FROM DARIO AMODEI ON OUR DISCUSSIONS WITH THE DEPARTMENT OF WAR

Anthropic believes in the importance of using AI to defend the US and other democracies and to defeat its autocratic adversaries. This is why it has worked proactively to deploy its models to the Department of War and the intelligence community. However, it also acknowledges that, in some cases, AI can undermine, rather than defend, democratic values. Two such use cases are mass domestic surveillance and full autonomous weapons. Anthropic has refused to accede to the Department of War's request to remove safeguards in its models. It will help the Department transition to another provider to avoid any disruption to ongoing military planning should it decide to offboard Anthropic.

XAI CO-FOUNDER TOBY POHLEN IS LATEST EXECUTIVE TO DEPART

xAI co-founder Toby Pohlen is leaving the startup. Pohlen was in charge of Macrohard, a division of the company focused on AI software run by agents. He is the seventh of 12 xAI co-founders to leave in less than three years. xAI merged with SpaceX earlier this month. SpaceX plans to go public in what is expected to be the largest initial public offering of all time.