Search the archive

FLOW FOR FREE

PROMPTFOO IS JOINING OPENAI

Promptfoo has agreed to be acquired by OpenAI. It will remain open source and continue to serve users and customers. The company was founded in 2024 to make it easy for developers to systematically test AI applications. It is joining OpenAI so that the security, evaluation, and compliance platform can have the greatest impact on how teams build and deploy AI. OpenAI will improve and integrate Promptfoo's core tech within its model and infrastructure layers so that teams can catch vulnerabilities early and ship secure AI from the beginning.

ANTHROPIC SUED US DEFENSE DEPARTMENT

Anthropic filed lawsuits in California and Washington, DC after the US Department of Defense labeled the company a supply-chain risk, which could bar government contractors from using its models. The dispute followed Anthropic's refusal to allow unrestricted military use of its AI systems, citing concerns over mass surveillance and fully autonomous weapons.

COPILOT COWORK: A NEW WAY OF GETTING WORK DONE

Microsoft's Copilot Cowork automates tasks across Microsoft 365, enhancing productivity through intelligent workflow management. Utilizing Work IQ, it coordinates emails, meetings, and files, performing actions like rescheduling and document preparation while maintaining user control. Currently in Research Preview, broader availability is expected in March 2026.

THE TOP 100 GEN AI CONSUMER APPS

Generative AI usage in consumer apps has evolved significantly, with products like CapCut, Canva, and Notion integrating AI as core features. ChatGPT remains the leading AI product globally, but competitors like Gemini and Claude are gaining traction, especially with US paid subscriptions. The AI landscape is diversifying geographically and functionally, with notable developments in video generation and agentic AI, notably OpenClaw, suggesting a shift beyond traditional browser and app metrics.

THE DEBT BENEATH THE DREAM

SoftBank's shares are deflating fast. Investors are more and more concerned that the company cannot pay its debts, and debtors are getting nervous. It is unclear how SoftBank will meet its commitments to OpenAI in April. Its borrowed money and illiquid assets mean that the firm's margin for error is far thinner than any announcement ever suggested.

HOW CODING AGENTS ARE RESHAPING ENGINEERING, PRODUCT, AND DESIGN

Coding agents have collapsed the traditional waterfall process by making implementation cheap and shifting the bottleneck from writing code to reviewing it. High throughput is the new constraint, favoring builders with product sense and systems thinkers who can manage rapid cycles. Generalists gain the most advantage by eliminating communication overhead, while the bar for pure specialization rises because domain depth alone is no longer enough to justify a role.

WHEN CODE IS FREE, RESEARCH IS ALL THAT MATTERS

As implementation costs collapse, the scarce resource becomes knowing what is worth building and whether it is feasible. Software engineering optimizes toward known solutions, but research operates without ground truth, making it resistant to the automation currently impacting traditional development. Research taste remains portable and nearly impossible to train, acting as the ultimate differentiator in a world where barriers to entry have dropped.

CODE REVIEW

Code Review for Claude Code allows developers to set up automated PR reviews that catch logic errors, security vulnerabilities, and regressions using multi-agent analysis. It analyzes GitHub pull requests and posts findings as inline comments. Findings are tagged by severity, and the tool doesn't approve or block PRs, allowing existing review workflows to stay intact. Claude Review is billed based on token usage, with reviews averaging $15 to $25.

TEACHING LLMS TO REASON LIKE BAYESIANS

Google researchers taught LLMs to reason in a Bayesian manner by training them to mimic the predictions of an optimal Bayesian model. Bayesian inference defines the optimal way to perform updates over multiple interactions. It could help LLMs optimize user interactions by updating their estimates of a user's preference. The researchers' approach significantly improves LLM performance on recommendation tasks and enables generalization to other tasks, suggesting the method taught LLMs to better approximate Bayesian reasoning.

PENGUIN‑VL: EFFICIENT VISION‑LANGUAGE MODELS

Tencent AI Lab released Penguin‑VL, a compact VLM family designed to improve multimodal efficiency by redesigning the vision encoder. It introduces Penguin‑Encoder, initialized from a text‑only LLM to align visual features with language representations and improve data‑efficient multimodal reasoning.

AI ASSISTANTS NOW EQUAL 56% OF GLOBAL SEARCH ENGINE VOLUME

AI assistants generate 45 billion monthly sessions, making up 56% of global search engine volume. Notably, 83% of AI usage occurs in mobile apps, with ChatGPT dominating at 89% of global AI sessions. Despite AI's rise, search engine activity remains crucial, with combined usage growing 26% globally since 2023.

NO, IT DOESN'T COST ANTHROPIC $5K PER CLAUDE CODE USER

A recent article by Forbes on Cursor claims that Anthropic's $200 per month Claude Code Max plan can consume $5,000 in compute. It is likely that the article's sources are confusing retail API prices with actual compute costs. The actual compute cost is roughly 10% of retail API prices. Most users don't come anywhere near their limits. Anthropic isn't a profitable company, but inference isn't why.

always know what's running. See a demo

EVERYONE USING AI HAS ABOUT 12 MONTHS TO DEVELOP THESE 3 MOATS

As AI defaults make products indistinguishable, the gap between lazy and intentional usage is closing.

TECHNOLOGICAL SPEED LIMIT

The learning rate of a technology has a maximum speed, and once you have enough money and talent, you saturate the learning rate - more simply doesn't help.

ORACLE IS BUILDING YESTERDAY'S DATA CENTERS WITH TOMORROW'S DEBT

Oracle faces setbacks with debt-fueled data center expansion as AI chip upgrades outpace construction timelines.

Get the free ebook

ANTHROPIC'S AI HACKED THE FIREFOX BROWSER. IT FOUND A LOT OF BUGS

Claude Opus 4.6 found more high-severity bugs in Firefox over a two-week period than the rest of the world typically reports in two months. The model discovered more than 100 bugs in total, 14 that were tagged as high severity. The model was asked to write code to exploit the bugs, but it turns out the model is much better at finding bugs than exploiting them. The exploits that Claude wrote would have been stopped in the real world by Firefox's other security mechanisms.

CLAUDE MARKETPLACE

The Claude Marketplace allows organizations to use some of their existing Anthropic commitment to purchase Claude-powered tools. Partner purchases will count against a portion of an organization's existing Anthropic commitment. Anthropic will manage all invoicing for partner spend. The marketplace is launching with GitLab, Harvey, Lovable, Replit, Rogo, and Snowflake, and Anthropic is looking for more companies to work with.

META SILENTLY LAUNCHES VIBES AI EDITOR

Meta silently launched the Vibes AI editor, transforming it from a feature in Meta AI to a standalone creation studio similar to Google Flow. The editor supports project creation, image and video generation, timeline editing, and other production tools. While the tooling is robust, output quality needs improvement.

THE EMERGING ROLE OF SRAM-CENTRIC CHIPS IN AI INFERENCE

SRAM-centric chips, like those from Cerebras and Groq, are gaining traction due to their advantages in AI inference workloads, particularly in minimizing latency and increasing throughput compared to traditional GPUs. This shift is driven by the demand for near-compute memory architectures, which offer faster data access than far-compute approaches like DRAM. The trade-off lies in balancing memory bandwidth and compute capacity, leading to new disaggregated hardware strategies that optimize both prefill and decode phases of AI tasks across varied hardware platforms.

REASONING BOOSTS SEARCH RELEVANCE 15-30%

Reasoning agents work best with simple search tools. Developers should use simple, easy-to-understand, and transparent search systems like grep or basic keyword search. Agents can sift through results, learn, and retry with what they've learned. Asking an agent to explain the intent behind a query helps the agent reason about how to best satisfy the request by forcing it to think about what the user wants.

YOUR LLM DOESN'T WRITE CORRECT CODE. IT WRITES PLAUSIBLE CODE.

A benchmarked LLM-generated Rust rewrite of SQLite ran 20,171x slower on primary key lookups because the query planner never checked the is_ipk flag, sending every WHERE clause through a full table scan instead of a B-tree search. The same author's disk cleanup daemon came in at 82,000 lines with a Bayesian scoring engine and PID controller to solve a problem that a one-line cron job already handles. The root failure is structural: LLMs optimize for plausible output matching the user's intent, and METR's RCT with 16 experienced open-source developers confirmed the problem scales, finding AI users were 19% slower while believing they were 20% faster.

methodology and results here.

KARPATHY'S AUTORESEARCH

Andrej Karpathy open-sourced autoresearch, a project for running AI-driven research loops on a small single-GPU LLM training setup. It lets agents modify code and guidance files, run short training experiments, evaluate results, and iteratively keep improvements overnight.

GOOGLE PM OPEN-SOURCES ALWAYS ON MEMORY AGENT, DITCHING VECTOR DATABASES FOR LLM-DRIVEN PERSISTENT MEMORY

The Always On Memory Agent is an agent system that ingests information continuously, consolidates it in the background, and then retrieves it later without relying on a conventional vector database. It is available on the official Google Cloud Platform GitHub page under a permissive MIT License that allows for commercial usage. Enterprise AI teams are moving beyond single-turn assistants into systems expected to remember preferences, preserve project context, and operate across longer horizons. The Always On Memory Agent offers a concrete starting point for that next layer of infrastructure.

CODEX SECURITY RESEARCH PREVIEW

OpenAI Codex Security is an application security agent designed to analyze repositories, identify high‑impact vulnerabilities, and suggest fixes.

CLAUDE FINDS 22 FIREFOX SECURITY VULNERABILITIES

OPENAI HARDWARE LEADER RESIGNED AFTER PENTAGON DEAL

OpenAI robotics and hardware executive Caitlin Kalinowski resigned following the company's reported partnership with the Pentagon. The departure highlighted internal tensions and broader debate around AI companies collaborating with defense organizations.

scraping headaches. Start with 250 free credits/month

IS THE AI COMPUTE CRUNCH HERE?

AI providers are struggling to meet soaring demand, with Anthropic degrading services due to unprecedented growth.

THE PROMPT I CANNOT READ

An LLM's output is shaped by an all-encompassing prompt, which includes system instructions, conversation history, and context files, forming the model's entire environment.

DO AI-ENABLED COMPANIES NEED FEWER PEOPLE?

AI startups operate with 40% smaller teams and are raising larger seed and series A rounds, indicating AI leads to more headcount efficiency.

SHAPING THE EXPLORATION OF THE MOTIVATION-SPACE MATTERS FOR AI SAFETY

Shaping motivation-space exploration in RL could enhance AI safety by preventing harmful misalignment.

Register for free →

GPT‑5.4

OpenAI released GPT‑5.4 and GPT‑5.4 Pro across ChatGPT, the API, and Codex. The model combines advances in reasoning, coding, and tool use to improve performance on professional workflows involving documents, spreadsheets, and software environments.

CHATGPT USERS RESEARCH PRODUCTS BUT WON'T BUY THERE, FORCING OPENAI TO RETHINK ITS COMMERCE STRATEGY

OpenAI is pulling back from processing purchases directly inside ChatGPT and will let app partners handle transactions instead. Purchases will now go through apps integrated into ChatGPT. OpenAI first announced its checkout feature in September 2025, but hardly any merchants signed up. While people are researching products on ChatGPT, they aren't following through with purchases there. The checkout process was a hands-on process. OpenAI had to onboard each active retailer individually, and it didn't have systems to collect and remit state sales taxes.

ANTHROPIC'S COMPUTE ADVANTAGE: WHY SILICON STRATEGY IS BECOMING AN AI MOAT

Anthropic has built the most diversified and cost-efficient compute architecture among frontier labs. OpenAI remains almost entirely dependent on Nvidia, and Microsoft's internal chip program is years behind schedule. Compute advantage amplifies model advantage. Anthropic's ability to deliver equivalent model quality at 30% to 60% lower cost per token is a compounding advantage on margin, training budget, and the pace of iteration.

LABOR MARKET IMPACTS OF AI: A NEW MEASURE AND EARLY EVIDENCE

This paper by Anthropic presents a new framework for understanding AI's impact on labor markets. Its goal is to establish an approach for measuring how AI is affecting employment. The approach aims to lay the groundwork to help future studies more reliably identify economic disruption. It is possible that the impacts of AI will be unmistakable, but the framework will help identify the more ambiguous effects.

FIVE AI VALUE MODELS FOR BUSINESS TRANSFORMATION

HOW GOOGLE AI UNDERSTANDS VISUAL SEARCH

Google detailed how advances in AI allow Lens and Circle to Search to identify and search multiple objects within a single image simultaneously. The update improves visual discovery by breaking down complex scenes like outfits or rooms into searchable components.