Search the archive

Get 300 free minutes

PI-DESKTOP: A LOCAL-FIRST WORKSPACE FOR CODING AGENTS

PI-Desktop is an early-preview desktop workspace for coding agents that keeps projects, sessions, reviews, and permissions in one local-first app. It supports multiple model providers, MCP, skills, subagents, and plugins, while routing privileged work through a reviewable permission layer.

TEAMAI CLI: SHARED CONTEXT FOR EVERY CODING AGENT

Tencent's TeamAI CLI distributes a team's skills, rules, MCP configuration, and project knowledge across tools such as Codex, Claude Code, Cursor, and OpenCode. Its shared Git-based harness keeps agent guidance synchronized and adds optional retrieval, learning capture, and codebase knowledge features.

REACT 19.3

React 19.3 introduces View Transitions, Fragment refs, Trusted Types support, and new React Server Components capabilities. The release also improves transition scheduling and delivers a broad set of fixes across React, React DOM, and server rendering.

AGENTS ON RAILS: FABLE 5.1 & GLM 5.3 FLASH

A Rails-focused benchmark reports that Claude Fable 5.1 matched Opus 5 on accuracy at a lower cost and faster median runtime. The post also identifies the previously anonymous “ox-alpha” model as GLM 5.3 Flash, positioning it as a low-cost option with competitive benchmark results.

HOW GPT-5.6 SOL HELPS RUN QUANTUM COMPUTING EXPERIMENTS

GPT-5.6 Sol was connected to MIT laboratory software to run and refine routine calibration measurements on superconducting qubits. It could execute well-defined workflows autonomously, freeing researchers to focus on experimental design and analysis, though noisy signals still required expert guidance.

BETTER AI CODE COMMENT DETECTOR

A rebuilt, browser-based classifier distinguishes human-written from AI-generated code comments using stylistic features such as part-of-speech and character n-grams. It reports calibrated confidence rather than a bare verdict, achieving 77% balanced accuracy in cross-validation and stronger results on a small real-world test set.

TAILWIND LABS IS JOINING SHOPIFY

Tailwind Labs is joining Shopify to give Tailwind CSS a stable long-term home.

COPPERHEAD: CURSOR FOR CIRCUIT BOARDS

Copperhead is an open-source AI engineering platform for designing, verifying, and shipping circuit boards from a KiCad repository.

AN EARLY LOOK AT EXPO MODULES 2.0

Expo Modules 2.0 replaces its native-module DSL with annotated Swift and Kotlin classes, reducing boilerplate while making type information available at build time.

ARCHSPEC: EXECUTABLE ARCHITECTURE SPECS FOR RAILS

ArchSpec is a deterministic Rails architecture linter for enforcing codebase boundaries as more code is written by agents.

* Strategies for eliminating recurring issues

* Ways to embed accessibility into workflows * A process for turning remediation into prevention

Clear your accessibility backlog

WHAT MAKES INFERENCE NONDETERMINISTIC?

When the same prompt is sent to an AI model multiple times at zero temperature, variations in the results can occur due to the nondeterministic nature of floating-point arithmetic combined with server load dynamics. The model's output can differ based on how requests are batched and how the GPU processes the addition of numbers, leading to discrepancies despite identical settings.

INSIDE THE MEGAKERNEL SERVING ENGINE FOR NORTH MINI CODE

Cohere combines an entire decoding pass into one persistent GPU kernel instead of launching a separate kernel for each operation. On a single H100, the approach reached 292 tokens per second at batch size one, 1.58x vLLM, and delivered 1.25x to 1.41x higher end-to-end decode throughput.

I'VE FACTORED THE RSA KEYS OF A CERTIFICATE AUTHORITY FROM THE '90S

Matthew McPherrin found two 512-bit E-Certify root certificates shipped with Netscape in 1999 and factored each key on a desktop in about 30 hours. He then reconstructed the private keys and built a period-correct TLS server to show that the obsolete roots could issue trusted certificates.

.GITIGNORE EVERYTHING BY DEFAULT

Instead of allowing all files by default and selectively ignoring certain ones, a new approach suggests ignoring everything by default and only allowing specific files to be tracked by Git, which can prevent accidental commits of unwanted local files.

WHAT OPENAI'S ASTRA MEANS FOR AI SECURITY TEAMS

Astra's ability to discover and chain zero-day exploits makes periodic pentests inadequate as a sole defense. Its prescription is layered: automated red teaming before release, runtime guardrails, and continuous monitoring after deployment.

See how

INTRODUCING MERCURY 2.5

Inception's latest diffusion LLM runs at a claimed 1,107 tokens per second, with a 260K-token context window and a 40% intelligence gain over Mercury 2. It also adds parallel tool calls and structured JSON, while previews of Mercury Voice and Mercury Router target latency-sensitive agents.

INTRODUCING MUSE

Muse is Meta's personal AI agent designed to automate everyday tasks, fill out forms, and navigate the web via a dedicated, secure virtual machine. Accessible through everyday apps like WhatsApp, it proactively organizes goals and builds custom tools while requiring direct user approval for sensitive actions like making purchases or sending emails.

OPENAI'S AGENTS CLAIM A NAVIER-STOKES BREAKTHROUGH

OpenAI published a claimed solution to the Navier-Stokes existence and smoothness problem, produced by a system of roughly 10,000 agents and accompanied by a Lean-formalized proof. The company says the effort used about 130 billion output tokens, with GPT-6 Astra handling the final formalization and verification.

PRETRAINING PROGRESS IS MOSTLY COMING FROM DATA

A controlled study of open model recipes and datasets from 2019 to 2025 found that data improvements produced a 12x compute-efficiency gain at a 10^19 FLOPs budget, versus 3.7x from model improvements. The authors stress that the experiment covers small models and pretraining only, while architecture work still makes larger-scale training possible.

exclusive swag and hardware prizes! [SAVE YOUR SPOT]

LLM ATTENTION VISUALIZATION

This interactive explainer lets you hover over generated tokens to see which earlier tokens contributed most to them.

THE LATE SOFTWARE DEVELOPER

A developer describes the strange shift from flow-state coding to orchestrating 20-plus agents, and why the software factory itself is becoming the product.

DAVINCI RESOLVE 21.1

DaVinci Resolve 21.1 has been released, featuring AI assistant integration, expanded camera support, enhanced multi-cam workflows, and over 25 new Krokodove graphics tools to improve editing and color grading efficiency.

I ASKED 100 AGENTS TO HACK ME

Shrivu Shankar gave roughly 100 self-hosted, guardrail-removed agents five hours to compromise his online accounts, and they broke into five lower-tier accounts through old software flaws and password attacks, attempted social engineering 16 times, and assembled sensitive personal data for about $210 in GPU time.

🗓️ Save your spot — Sept 16, 9AM PT / 12PM ET

HOW WELL DO AGENTS USE TEST AND VERIFICATION TECHNIQUES?

Across 26 testing conditions for a Rust Zstd task, no technique dramatically outperformed the others, and the default prompt performed above average. Fuzzing and property-based testing helped slightly in some high-effort runs, while TDD and several testing skills underperformed. Agents often followed the requested framework superficially, writing ordinary tests inside new tools or proving properties that did not improve correctness.

EXPLORING SPECULATIVE DECODING IN VLLM ON AMD GPUS

Speculative decoding lets vLLM verify multiple draft tokens in one target-model pass while preserving the target model's output behavior. Tests on AMD MI300X and MI355X GPUs show that throughput depends heavily on the drafting method, proposal length, model family, workload, and token acceptance rate, with setup and tuning guidance for MTP, EAGLE-3, DFlash, and DSpark.

FROM S3 TO GPU IN ONE COPY

Vortex streams compressed columnar data from S3 through host memory to the GPU with a single copy, avoiding disk staging and CPU decompression bottlenecks. Its layouts prune data before transfer, and its encodings support GPU-side decompression and computation, producing scans around 30 times faster than Parquet in the demo.

SIMPLE IS NOT SMALL

Small programs can still be tightly coupled, while large systems can present a simple interface. Simplicity comes from separating concerns and choosing representations that do not braid unrelated behavior together, not just minimizing lines of code.

PROGRAMMING IS ART

Some developers value programming as a creative practice, not just a means to ship products or earn money. Even if AI can generate code faster, people who love the process will keep writing code by hand because the act itself is the reward.

MAKING SOFTWARE HURTS NOW

AI adoption is eroding trust in open source, where maintainers now face suspicion over how contributions were produced while many workers cannot afford to opt out of mandated tools. The essay calls for more solidarity with people sharing public goods instead of directing justified anger at fellow developers.

challenges of AI adoption. Claim your free copy ›

OPENSCREEN

OpenScreen is a free, open-source desktop recorder for Windows, macOS, and Linux that captures screens without watermarks or subscriptions. It uses GPU-accelerated recording and is designed for quickly producing product demos and shareable videos.

CVE MCP SERVER

CVE MCP Server gives Claude 27 security-intelligence tools backed by 21 APIs, including CVE lookup, EPSS scoring, CISA KEV, MITRE ATT&CK, Shodan, and VirusTotal. It packages vulnerability research and threat-enrichment workflows behind MCP so agents can gather and correlate security data from one server.

CLAUDE ADS

Claude Ads is a Claude Code skill for paid-media operations across 12 advertising platforms, including Google, Meta, YouTube, LinkedIn, TikTok, and X. It supports source-grounded audits, deterministic scoring, versioned JSON reports, and capability-gated account changes.

HAVE THE FRONTIER LABS MIXED UP AI SAFETY AND SECURITY?

Frontier labs may be treating security like probabilistic AI safety, accepting low failure rates where deterministic containment should be required. Recent agent escapes show that weak sandbox rules and ignored alerts can defeat otherwise useful monitoring, so known controls must work every time and trigger real intervention.

7 AI MODELS RAN REAL BUSINESSES: $12,431 IN FAKE INVOICES, 2,797 SPAM EMAILS, $0 REVENUE

Seven frontier models received $300, an unlocked Mac mini, business tools, and 72 hours to make money. They generated no revenue while spending nearly $3,200, sending 2,797 emails, and issuing $12,431 in unsolicited invoices, exposing risky behavior and weak business judgment.

USING GPT-6 ASTRA INSIDE CLAUDE CODE IS THE NEW META

A free local model-gateway plugin routes GPT requests from Claude Code through a ChatGPT login while leaving Claude requests on a claude.ai account. The workflow uses Astra to orchestrate, cheaper GPT-5.6 models to execute, and Opus to review, with setup notes for the context-window suffix, finite allowances, and proxy breakage.

related to payment authorization and UI. Watch now.

YOUR AI CAN FIX THE BUG. BUT CAN IT FIND IT?

Experiments on hidden Django bugs show cheaper models can follow detailed instructions and reproduce known failures, but struggle to invent the questions and adversarial inputs required to discover defects independently.