Blog

AI is accelerating into production — new models, hosted agents, developer tools, massive investments and cloud deals slash code costs and boost app creation. Simultaneously, safety, security, workforce and intervention risks — from exploit-prone models to regulatory labels and harmful well-intentioned programs — demand stronger oversight.

• WSJ: Meta Announces New AI Model ‘Muse Spark’ (Apr. 8, 2026)
  Meta unveiled Muse Spark, a closed large language model to power its AI chatbot, with a private API preview for partners. Meta says Muse Spark rivals Gemini, OpenAI, and Anthropic.
• Simon Willison: Meta’s new model is Muse Spark, and meta.ai chat has some interesting tools (Apr. 8, 2026)
  Meta released Muse Spark, a hosted model on meta.ai in a private preview, with Instant and Thinking modes. Its chat offers web and Meta search, image generation, a Python container with visual grounding, and subagents for sandboxed analysis.
• Claude: Claude Managed Agents: get to production 10x faster (Apr. 8, 2026)
  Claude Managed Agents launches public beta of composable APIs that run cloud-hosted agents with secure sandboxing, long sessions, and multi-agent coordination. 
• Simon Willison: Anthropic’s Project Glasswing sounds necessary to me (Apr. 7, 2026)
  Anthropic withheld Claude Mythos, giving preview access to trusted Project Glasswing partners, citing its powerful, exploit-finding capabilities. Mythos autonomously found and chained thousands of high-severity vulnerabilities.
• WSJ: AI Giants Go on Charm Offensive to Avert Public Backlash (Apr. 7, 2026)
  Big AI companies are shifting to reassure a worried public, offering policy ideas, partnerships, and training to manage job losses, wealth gaps, and other harms.
• WSJ: Amazon CEO Andy Jassy Makes Case for Big AI Spending in Annual Letter (Apr. 9, 2026)
  Andy Jassy says Amazon will invest heavily in AI, robotics, and satellite internet, pledging $200 billion in 2026 capex for AI infrastructure, custom chips, and Leo satellites.
• WSJ: CoreWeave Secures $21 Billion Expanded AI Cloud Deal Agreement With Meta (Apr. 9, 2026)
  CoreWeave expanded its long-term AI cloud deal with Meta through December 2032 for about $21 billion.
• 9to5Mac: App Store sees 84% surge in new apps as AI coding tools take off (Apr. 6, 2026)
  AI coding tools have driven a 30% surge in App Store submissions to nearly 600,000, enabling nonprogrammers and developers to build apps quickly.
• Pere Villega: Code Is Cheap Now, And That Changes Everything (Mar. 15, 2026)
  AI coding tools have slashed the cost and time of building software, letting one person replace teams and months. 
• Alex Tabarrok: AI, Unemployment and Work (Apr. 9, 2026)
  AI-driven automation could cut total work hours, which might mean mass unemployment or a shorter workweek, depending on whether gains are taxed, shared as dividends, or converted into holidays.
• NY Times: Federal Court Denies Anthropic’s Motion to Lift ‘Supply Chain Risk’ Label (Apr. 8, 2026)
  A federal appeals panel denied Anthropic’s request to block the Defense Department from labeling it a supply-chain security risk, barring new Pentagon contracts.
• The Atlantic: These Teens Got Therapy. Then They Got Worse. (Nov. 6, 2023)
  An Australian trial that taught a shortened DBT program to over 1,000 middle-schoolers produced worse short-term outcomes, including higher anxiety, depression, and poorer parent relationships, some effects persisting six months later. Universal, nonconsensual programs may backfire, especially without parental involvement.

AI commercialization: tech firms race to deploy models, invest in compute, chips and partnerships to scale products. Societal risks: AI boosts productivity but drives job disruption, security, legal and military vulnerabilities, demanding urgent governance.

• WSJ: Meta Announces New AI Model ‘Muse Spark’ (Apr. 8, 2026)
  Meta released Muse Spark, a closed large language model that will power its AI chatbot, with a private API preview for partners, developers, and possible later open-source versions.
• WSJ: An Inside Look at OpenAI and Anthropic’s Finances Ahead of Their IPOs (Apr. 5, 2026)
  OpenAI and Anthropic are racing to IPOs, speeding model releases, and pouring huge sums into AI training. OpenAI predicts $121 billion in 2028 computing costs and $85 billion in losses, while Anthropic expects smaller, but rising, costs.
• WSJ: Intel Partners With SpaceX, Tesla to Operate New Chip Plant (Apr. 7, 2026)
  Intel will partner with SpaceX, xAI, and Tesla to design, fabricate, and package ultra-high-performance chips at the Terafab plant in Texas.
• WSJ: The Workers Opting to Retire Instead of Taking on AI (Apr. 6, 2026)
  Some older workers are retiring early because AI is changing jobs, adding stress, and challenging autonomy, while rising home equity and stock gains make it affordable.
• Tyler Cowen: Andy Hall advice on AI and economic research (Apr. 7, 2026)
  An experiment showed Claude Code could extend old papers—planning, scraping, coding, running regressions, and drafting results in about 45 minutes, though it made mistakes.
• Simon Willison: Eight years of wanting, three months of building with AI (Apr. 5, 2026)
  A project used AI to rapidly build a high-fidelity SQLite devtools suite, producing a fast prototype in months after years of thinking. AI sped coding, but muddled architecture, forcing human-led redesigns to make a robust, maintainable, long-lived library.
• Alex Tabarrok: AI Risks (Apr. 8, 2026)
  Two papers warn of opposite AI risks. One shows a model that finds thousands of severe software flaws across major systems, browsers, and services; the other shows web agents can be hijacked via hidden text, steganography, and embedded commands.
• WSJ: A Judge Mistakes the Claude Chatbot for a Person (Apr. 6, 2026)
  A Manhattan judge ruled a defendant waived attorney–client privilege by using the Claude AI, letting prosecutors seize his prompts, outputs, and logs.
• WSJ: Five Risks Jamie Dimon Is Worried About in 2026 (Apr. 6, 2026)
  Jamie Dimon warns of AI’s job disruption, and a weak, divided Europe.
• WSJ: Elon Musk’s Starship Heavy Could Revolutionize Warfare (Apr. 5, 2026)
  Super-cheap, high-capacity Starship launches could let militaries stockpile, arm, and drop thousands of conventional munitions or drones from low Earth orbit, enabling near-instant strikes anywhere on Earth.

AI is rapidly automating and accelerating work—enabling agent-driven developer workflows, faster security testing, prototype-first product cycles, and even rapid business creation. Simultaneously, widespread deployment raises ethics, accuracy, and governance concerns—from flawed "AI detector" false positives to surveillance in education—requiring robust human oversight.

• Claude: Extra usage credit for Pro, Max, and Team plans (Apr. 3, 2026)
  Pro, Max, and Team subscribers can claim a one-time usage credit equal to their subscription price from April 3–17, 2026.
• WSJ: OpenAI’s Top Executive Fidji Simo to Take Medical Leave From Company (Apr. 3, 2026)
  OpenAI product chief Fidji Simo is taking weeks of medical leave to treat a worsening neuroimmune condition.
• Cursor: Meet the new Cursor (Apr. 1, 2026)
  Cursor 3 is an agent-first workspace that unifies local and cloud agents, supports multi-repo work, and runs agents in parallel.
• OpenAI: OpenAI acquires TBPN (Mar. 31, 2026)
  OpenAI to bring its editorial team, audience insights, and communications expertise to company.
• Skyvern Blog: Getting Claude to QA its own work (Apr. 3, 2026)
  Skyvern added browser tools, Claude Code skills, and CI smoke tests that read git diffs, generate tests, run browser interactions, and post pass/fail results with screenshots.
• ChinaTalk: China's AI Education Experiment – by Lily Ottinger (Apr. 3, 2026)
  China is pushing AI into K–12 schools to grade art, scan faces, screen mental health, and provide basic AI access by 2030. 
• NY Times: How A.I. Helped One Man (and His Brother) Build a $1.8 Billion Company (Apr. 2, 2026)
  Matthew Gallagher built Medvi, a telehealth GLP‑1 seller, using A.I. to code, market, and run operations after $20,000 and two months. With just his brother, it made $401 million in year one, and is on track for $1.8 billion.
• Business Insider: Jack Dorsey says Block employees now bring prototypes — not slide decks — to meetings (Apr. 2, 2026)
  Tech leaders are ditching slide decks, favoring prototypes, memos, and interactive demos that show real or simulated data, can be modified in real time, and cut the cost of being wrong. Jack Dorsey says Block now brings prototypes to meetings.
• The Register: Amazon security boss: AI makes pentesting 40% more efficient (Apr. 1, 2026)
  Amazon reports over 40% efficiency gains using AI to automate pentesting before and after launch. AI handles mundane work, finds and chains vulnerabilities, and alerts humans for decisions.
• WSJ: The ‘AI Detector’ as Defamation Machine (Apr. 3, 2026)
  An AI tool, Pangram, flagged novels, op-eds, and columns as AI-generated, producing inconsistent, sometimes contradictory results and raising false-positive concerns. Its study, partly authored by Pangram staff, lacked author outreach, creating conflicts of interest, and risking unfair accusations.

Two themes: (1) commercialization and scaling strains—AI products face quotas, ad-driven monetization, massive compute costs, shutdowns, and huge data‑center builds; (2) capability versus safety—agents enable powerful automation (useful and risky), but fragile harnesses demand stronger engineering, governance, and public scrutiny.

• The Register: Anthropic admits Claude Code quotas running out too fast (Mar. 31, 2026)
  Claude Code users report rapid quota exhaustion, disrupting work, and Anthropic is investigating.
• Neowin: Microsoft Copilot is now injecting ads into pull requests on GitHub (Mar. 31, 2026)
  Microsoft’s AI tools are injecting ads into pull requests via hidden Copilot comments, promoting services like Raycast, Slack, Teams, and IDEs.
• Claude: Computer use is now in Claude Code. (Mar. 30, 2026)
  Claude Code can open apps, click through user interfaces, and test what it builds, all from the command line.
• Axios: Everyone's worried that AI's newest models are a hacker's dream weapon (Mar. 29, 2026)
  Anthropic’s unreleased model, Mythos, reportedly lets agentic AI autonomously plan and execute large-scale cyberattacks, making such strikes far more likely.
• WSJ: Sayonara, Sora: OpenAI Says Fun Time Is Over (Mar. 29, 2026)
  OpenAI shut down Sora after massive compute costs and chip shortages forced a move from free spectacles to revenue-generating uses.
• WSJ: The Sudden Fall of OpenAI’s Most Hyped Product Since ChatGPT (Mar. 29, 2026)
  OpenAI shut Sora, its consumer video app backed by Disney, after it proved unprofitable and consumed too many AI chips.
• WSJ: Ex-BP CEO Bernard Looney to Lead Data-Center Developer (Mar. 31, 2026)
  Bernard Looney, former BP chief, is taking the CEO role at Prometheus Hyperscale to lead massive, AI-focused data-center projects in Wyoming and Texas, aiming sustainable, liquid-cooled facilities. 
• NY Times: When Will Japan’s Cherry Blossoms Bloom? A.I. Can Help Answer That (Mar. 30, 2026)
  AI helps Japanese forecasters analyze decades of weather data and thousands of crowdsourced photos to predict cherry blossom timing more precisely.
• Simon Willison: A quote from Georgi Gerganov (Mar. 30, 2026)
  Local models often fail not because of the model itself, but due to fragile harnesses, chat templates, prompt construction, and inference bugs. 
• Center for Humane Technology: “The AI Doc” Opens Friday. It's The Film We Need In This Moment. (Mar. 26, 2026)
  The AI Doc opens in U.S. theaters, showing AI’s promise, peril, and power in a shared, urgent experience. It calls for joining the Human Movement, using the discussion guide, and acting together for a humane AI future.
• Microsoft Copilot: Copilot – Terms of Use (Mar. 31, 2026)
  Microsoft’s legal team is clearly at work: “Copilot is for entertainment purposes only. It can make mistakes, and it may not work as intended. Don’t rely on Copilot for important advice. Use Copilot at your own risk.”
• WSJ: Buying the Dip? This AI Agent Will Do It for You (Mar. 31, 2026)
  Public is rolling out AI agents to automate investing, execute trades, and apply orders like stop-losses, with users setting, reviewing, and pausing workflows. The tool passed Series 7 exams.
• The Fine Print: "CEO Said A Thing!" Journalism (Mar. 30, 2026)
  US media often parrots CEOs without context or challenge, amplifying false, self-serving claims. This “CEO said a thing!” coverage props up corporate power, normalizes myths about tech leaders, and blurs news with marketing.