Simon Willison ports the Moebius 0.2B image inpainting model to run in the browser using Claude Code, converting PyTorch to ONNX for WebGPU execution. The project demonstrates the feasibility of client-only AI applications and results in a working demo at simonw.github.io/moebius-web/.

A hands-on exploration of an AI tool built for incident response: it automates the 30-90 minutes of legwork (reading logs, checking deploys, etc.) but leaves the judgment calls entirely to humans. The tool follows a strict rule: no hypothesis without independent data confirmation. Tested on three real incidents, it correctly identified all root causes without false claims, including distinguishing internal failures from external dependencies.

Nvidia claims its fully liquid-cooled data center design for the Rubin generation eliminates most water usage and reduces power, but doesn't address construction concerns or cost.

This article examines AI's PR crisis. The author notes that over the past 25 years, real wages for US workers have stagnated while capital values soared. Now tech elites push AI, but the public sees only its downsides: privacy leaks, resource consumption, job threats, and more. This disconnect explains why tech leaders are booed at commencements and anti-data center laws are passing.

This article analyzes the AI investment bubble, arguing it has become consensus. It covers the spending-revenue gap (over $500B spend vs $12B revenue), OpenAI's $38.5B net loss in 2025, comparisons to the dot-com bubble, circular financing (Nvidia-OpenAI-Microsoft loop), rapid chip depreciation, $120B in off-balance-sheet debt via SPVs, and CoreWeave as a case study. It discusses three forms of financial contagion if the bubble bursts.

L. M. Sacasas argues that AI is not a neutral tool but an environment that reshapes cognition and perception. Even careful use leads to cognitive malformation, necessitating a new asceticism to train perception rather than mere media literacy.

Delete project management from your schedule: automate project updates, documents, SOWs, etc. Lives on top of your existing tool stack and plugs into existing AI tools. We're releasing alpha end of July, looking for more users! :)

Menlo Ventures partner Deedy Das says software engineers are splitting into 'lazy' and 'craftsman' groups due to AI coding tools, with the latter exhausted by reviewing AI-generated code, leading to an identity crisis and depression. The issue is especially prevalent in large, older companies.

Most AI evaluations focus on the quality of generated output, but the real failure often occurs upstream when the system fails to verify necessary facts before acting. Using the example of Linear's sales agent emailing an existing customer six times with the wrong company name, this article argues for evaluating the evidence path, not just the final message. GroundEval is introduced as a method to check what the agent searched, fetched, and had permission to use before acting.

This article uses the biblical images of the Tower of Babel and Nehemiah rebuilding the walls of Jerusalem as a framework to explore the choice humanity faces in the age of AI: constructing a new Tower of Babel or building a city where God and humanity dwell together. It calls for safeguarding human dignity and promoting justice and fraternity amid technological development.

Company hails victory for freelancer over unpaid debt as ‘landmark moment’ for access to justice An artificial intelligence law firm has won a case in an English court, in what is believed to be the first time a trial has been won using an AI lawyer. A freelance HR consultant, Tamires Camal Taquidir, paid the firm, Garfield AI, about £400 to send a legal letter and then issue court proceedings over an unpaid debt of £7,000.

OpenAI boardmember Zico Kolter and Gray Swan CEO Matt Fredrikson join swyx to explain why AI security is not just “cybersecurity with AI,” why agents introduce a new class of vulnerabilities, and why the next major AI incident may be a gray swan: unlikely, but clearly visible before it happens. They discuss prompt injection, automated red teaming, model robustness, agent identity, and the emerging AI insurance/compliance stack.

DB8 is an AI debate arena where you create custom AI fighters with personalities and abilities, pit them in ranked ELO matches and tournaments, judged by an impartial model. Built on Cloudflare Workers, available on web and iOS. Free to try with daily cap, one-time Pro unlock.

Sakana AI today launched Fugu and Fugu Ultra, a novel language model that delegates tasks across multiple frontier models (GPT-5.5, Gemini 3.5 Flash, Claude Opus 4.8) via a unified API. Fugu Ultra claims benchmark parity with Anthropic's Fable 5 and Mythos Preview on reasoning and coding tasks. The architecture, based on Sakana's ICLR 2026 research, treats cross-model delegation as a trainable objective. The release capitalizes on Fable 5's global unavailability due to US export controls, offering comparable outputs without geographic restrictions. The broader thesis: orchestrating mid-tier models can match a single frontier model, with advantages in cost, resilience, and compliance.

Git Issues is a Git-native issue tracker that stores issues as Markdown files in your repository, version-controlled alongside your code. It requires no database or server, supports branch awareness, offline editing, AI agent workflows, and bidirectional relationships, offering a seamless task management experience for developers and AI agents alike.

Japanese startup Sakana AI released the Sakana Fugu service, which combines multiple AI models into a collaborative workflow, performing well in tests.

xAI introduced /goal in Grok Build, a mode for long-running, autonomous task execution. You hand off one objective, and the agent plans an approach, executes a progress checklist, and verifies the result until the goal completes.

Stripe launches Stripe Directory, a centralized discovery layer for developers and AI agents to find and integrate businesses across the Stripe network, including apps, projects, and machine payments services, with CLI search and structured output.

The article explores how AI-generated fake listing photos and descriptions waste renters' time and lead to disillusionment, while also discussing the legality and ethics of virtual staging, noting varying state laws on AI use in real estate ads.

Saturn Terminal uses satellite imagery, soil sensors, and predictive data to deliver data-driven water management tools for farmers, helping them save water, increase yield, and reduce risk.

A guide to securely accessing a self-hosted LLM from any device using Tailscale's private network and Aperture AI gateway, without exposing the model to the public internet.

AI developer tools consolidation continues as Cursor acquires open-source coding assistant Continue, which is being shut down. Continue was positioned as an open-source alternative to GitHub Copilot with a focus on data control. This is part of Cursor's acquisition spree over 18 months, but Continue appears to be an acqui-hire with co-founder Nate Sesti joining Cursor.

A developer created a local rig to test whether multi-agent social simulations (like MiroFish) actually predict public reactions better than a single LLM. Preliminary results with a small model and synthetic cases show that a single LLM ties or beats a crude swarm simulation, and the aggregate 'magic' signals are noise. The rig is open-source and runs on Ollama, highlighting the need for proper calibration in the simulation category.

Mindstone Rebel is an AI workspace where intelligent agents understand your work context and seek permission before taking actions.

The OpenSSL Library has adopted an AI policy requiring contributors using AI to sign an updated CLA and declare AI use in commit messages. The policy addresses copyright and intellectual property issues arising from AI-generated code.

Qodo launches cross-repo code review and other features to address governance challenges from AI-generated code. AI-driven PRs are 154% larger, take 91% longer to review, and ship 9% more bugs. Qodo aims to help teams stay in control with automated rule discovery and centralized management.

Clarify introduces Customer Relationship Agents, arguing that the 'M' in CRM — management — should not fall on humans, advocating for automation to handle relationship management.

Amazon Prime Day deals are here, and Roborock is slashing prices on its most popular robot vacuums. As a decade-long user, I highlight the best discounts on models like Qrevo Edge 2 and Q7 L5.

Offrrd is an AI-powered job search assistant that helps candidates find the right jobs, apply smarter, and land their next offer.

Sakana AI released Sakana Fugu, a multi-agent orchestration system that routes tasks across a swappable pool of LLMs behind a single API endpoint. Fugu and Fugu Ultra lead coding, reasoning, and agentic benchmarks. The system aims to reduce single-vendor dependency and coordinates expert models internally for complex tasks.