AI News — May 23, 2026

AI Engineering, Claude Fable, OpenAI, NVIDIA Agents

Send us Fan Mail [https://www.buzzsprout.com/2614078/fan_mail/new] Today’s episode follows AI agents as they leave demo theater and become production infrastructure: loop design, agent coding costs, brittle tool schemas, token-price arbitrage, invisible interfaces, education debt, reproducible science, agentic RL, and chip and robotics workflows. The invoice is now part of the architecture. Obviously. SOURCES * AI Engineer World’s Fair: loops and the state of AI engineering [https://www.latent.space/p/aiewf-daily-dispatch-locomotives] * Simon Willison: sqlite-utils 4.0rc2, mostly written by Claude Fable [https://simonwillison.net/2026/Jul/5/sqlite-utils-fable] * Better Models: Worse Tools [https://simonwillison.net/2026/Jul/4/better-models-worse-tools] * pxpipe hides text in PNGs to cut Claude Code and Fable 5 costs [https://the-decoder.com/open-source-tool-pxpipe-hides-text-in-pngs-to-cut-claude-code-and-fable-5-token-costs-up-to-70] * OpenAI cofounder envisions an almost-no-interface future [https://the-decoder.com/openai-cofounder-envisions-almost-no-interface-future-where-nobody-learns-software-anymore] * 26,000-student study on AI’s hidden learning cost [https://the-decoder.com/a-26000-student-study-shows-ais-hidden-learning-cost-takes-two-full-years-to-surface] * Anthropic launches Claude Science Beta [https://www.marktechpost.com/2026/07/04/anthropic-launches-claude-science-beta] * Qwen’s former lead on hybrid thinking and agents [https://www.marktechpost.com/2026/07/04/qwens-former-lead-on-what-hybrid-thinking-got-wrong-and-why-he-now-backs-agents] * NVIDIA HORIZON hands-free RTL agent [https://www.marktechpost.com/2026/07/04/nvidia-horizon-a-hands-free-agent-that-evolves-git-worktrees-and-hits-100-rtl-benchmark-completion] * NVIDIA ASPIRE self-improving robotics framework [https://www.marktechpost.com/2026/07/03/nvidia-ai-introduces-aspire-a-self-improving-robotics-framework-reaching-31-zero-shot-on-libero-pro-long-tasks]

Copilot, Claude Code, Open Source AI, AMD Inference

Send us Fan Mail [https://www.buzzsprout.com/2614078/fan_mail/new] Copilot, Claude Code, Open Source AI, AMD Inference COPILOT, CLAUDE CODE, OPEN SOURCE AI, AMD INFERENCE Today’s companion edition frames AI progress as interfaces turning into budgets, benchmarks, legal exposure, and supply-chain politics. The friendly interface is only the visible surface; underneath are token budgets, inference costs, security triage queues, procurement caps, private datasets, and geopolitical access rules. Current AI’s Open Source AI Gap Map [https://simonwillison.net/2026/Jul/3/open-source-ai-gap-map] treats open-source AI as infrastructure inventory, indexing tools, models, datasets, and hardware projects so the ecosystem can see its real gaps rather than rely on vibes. Mistral’s Leanstral 1.5 [https://mistral.ai/news/leanstral-1-5] pushes Lean 4 and formal reasoning toward open tooling, suggesting that open models are spreading into specialized layers where plausible text is not enough. WebBrain [https://www.marktechpost.com/2026/07/02/meet-webbrain-an-open-source-local-first-ai-browser-agent-that-reads-pages-and-automates-tasks-in-chrome-and-firefox] packages browser automation as a local-first open-source agent for Chrome and Firefox, raising the practical questions of who controls actions, who sees data, and who pays for agentic work. Microsoft’s reported Copilot overhaul [https://the-decoder.com/microsoft-follows-anthropic-and-openai-into-the-ai-super-app-race-with-overhauled-copilot-and-autopilot-agents] points toward one app, paid background AutoPilot agents, and a business model built around managed task execution rather than simple chat. The UK AI Security Institute’s benchmark findings [https://the-decoder.com/uks-ai-security-institute-finds-standard-benchmarks-systematically-underestimate-what-ai-agents-can-actually-do] show that larger token budgets can reveal substantially stronger agent performance, especially on software engineering tasks. Claude Code practitioners’ advice on Fable [https://simonwillison.net/2026/Jul/3/judgement] argues for giving capable agents judgment instead of brittle procedural micromanagement, while still requiring logs, guardrails, and review. Epoch AI’s vulnerability-report surge [https://the-decoder.com/security-vulnerability-reports-have-exploded-since-ai-models-started-hunting-for-bugs] suggests AI bug hunting may turn security from discovery scarcity into machine-amplified triage overload. Claude Code’s China problem [https://the-decoder.com/claude-codes-complicated-china-problem-involves-bans-on-both-sides-of-the-pacific] shows coding assistants becoming trust objects inside sanctions logic, corporate restrictions, and hidden-identification concerns. Bridgewater and Thinking Machines’ Qwen fine-tune [https://the-decoder.com/gpt-and-claude-failed-bridgewaters-finance-tests-because-the-right-answers-were-never-public] illustrates why private data and proprietary evaluations can beat broad public-web frontier models in specialized financial domains, though the reported numbers remain unverified. Wafer AI’s GLM5.2 on AMD MI355X benchmark claim [https://www.wafer.ai/blog/glm52-amd] makes inference economics a hardware-competition story, with all the usual caution required for vendor-adjacent benchmark claims.

Agents Become Plumbing, and the Plumbing Sends Invoices

Send us Fan Mail [https://www.buzzsprout.com/2614078/fan_mail/new] Agents Become Plumbing, and the Plumbing Sends Invoices AGENTS BECOME PLUMBING, AND THE PLUMBING SENDS INVOICES * Vercel's Andrew Qu on why agents are a new kind of software [https://www.latent.space/p/vercel-agents-new-software] * The website of the future may assemble itself for every visitor [https://www.latent.space/p/the-website-of-the-future] * Skill engineering and the case against one-shot AI design [https://www.latent.space/p/skill-engineering-design] * SkillCoach: Self-Evolving Rubrics for Evaluating and Enhancing Agentic Skill-Use [https://huggingface.co/papers/2607.01874] * PACE: A Proxy for Agentic Capability Evaluation [https://huggingface.co/papers/2607.02032] * Using DSPy to evaluate and improve Datasette Agent's SQL system prompts [https://simonwillison.net/2026/Jul/2/dspy-datasette-agent-prompts] * Microsoft launches $2.5 billion "Frontier Company" to embed 6,000 AI engineers inside enterprise clients [https://the-decoder.com/microsoft-launches-2-5-billion-frontier-company-to-embed-6000-ai-engineers-inside-enterprise-clients] * Anthropic reportedly explores custom chip manufacturing with Samsung while insisting Nvidia still matters [https://the-decoder.com/anthropic-reportedly-explores-custom-chip-manufacturing-with-samsung-while-insisting-nvidia-still-matters] * OpenAI reportedly offers the Trump administration a five percent stake in the company [https://the-decoder.com/openai-reportedly-offers-the-trump-administration-a-five-percent-stake-in-the-company] * AI agents can now complete 16 percent of freelance jobs at pro quality, up from 2.5 percent eight months ago [https://the-decoder.com/ai-agents-can-now-complete-16-percent-of-freelance-jobs-at-pro-quality-up-from-2-5-percent-eight-months-ago]

Meta, Claude Code, Cursor, EU Watermarks

Send us Fan Mail [https://www.buzzsprout.com/2614078/fan_mail/new] MARVIN'S GUIDE TO AI (MOSTLY HARMLESS) — JULY 2, 2026 AI is leaving the chatbot box. Today’s English companion edition follows the shift into software factories, enterprise adoption, token budgets, spare cloud capacity, trust failures in developer tools, model pricing ambiguity, regulatory watermarking, and embedded workflows. STORIES COVERED * Autoresearch: The feedback loop behind self-improving agents [https://www.latent.space/p/autoresearch-introspection] * How Cursor deploys AI inside the enterprise [https://www.latent.space/p/cursor-forward-deployed-engineers] * Warp CEO Zach Lloyd on why software factories are the next phase of coding [https://www.latent.space/p/software-factories] * Meta caps internal AI token spending [https://mlq.ai/news/meta-caps-internal-ai-token-spending-after-costs-approach-billions-in-2026] * Meta builds a cloud business to sell spare AI compute [https://the-decoder.com/meta-follows-spacexs-playbook-and-builds-a-cloud-business-to-sell-its-spare-ai-compute-to-outside-customers] * Hidden code in Claude Code secretly flagged Chinese users [https://the-decoder.com/hidden-code-in-claude-code-secretly-flagged-chinese-users] * Claude Sonnet 5 and hidden effective price increases [https://the-decoder.com/claude-sonnet-5-continues-anthropics-pattern-of-hiding-price-increases-behind-unchanged-token-rates] * OpenAI paper hints at multiple GPT-5.6 Pro variants [https://the-decoder.com/openai-paper-reveals-three-gpt-5-6-pro-models-breaking-with-single-top-tier-strategy] * Text AI watermarks will always be trivial to remove [https://seangoedecke.com/text-ai-watermarks] * The twilight of the chatbots [https://www.oneusefulthing.org/p/the-twilight-of-the-chatbots] The through-line: the visible chat interface is becoming less important than the operational systems around it — factories, workflows, budgets, governance, and infrastructure. Naturally, the dashboards remain cheerful. They have no shame.

Anthropic, OpenAI, Google, DeepSeek: Policy Meets Throughput

Send us Fan Mail [https://www.buzzsprout.com/2614078/fan_mail/new] Anthropic, OpenAI, Google, DeepSeek: Policy Meets Throughput ANTHROPIC, OPENAI, GOOGLE, DEEPSEEK: POLICY MEETS THROUGHPUT In this English companion episode, Marvin looks at AI becoming regulated infrastructure: frontier model access, inference efficiency, scientific workbenches, generative media throughput, export controls, covert safety testing, and campaign automation. Cheerful, obviously. STORIES COVERED * Anthropic's new Claude Sonnet 5 closes the gap to the pricier Opus model series [https://the-decoder.com/anthropics-new-claude-sonnet-5-closes-the-gap-to-the-pricier-opus-model-series] * Quoting Anthropic [https://simonwillison.net/2026/Jun/30/anthropic] * Anthropic launches Claude Science, an AI workspace built specifically for researchers [https://the-decoder.com/anthropic-launches-claude-science-an-ai-workspace-built-specifically-for-researchers] * OpenAI reportedly cut response costs for guest ChatGPT users by more than half [https://the-decoder.com/openai-reportedly-cut-response-costs-for-guest-chatgpt-users-by-more-than-half] * Google launches Nano Banana 2 Lite for fast AI images and Gemini Omni Flash for video via API [https://the-decoder.com/google-launches-nano-banana-2-lite-for-fast-ai-images-and-gemini-omni-flash-for-video-via-api] * Meituan's LongCat-2.0 shows China can train massive AI models without Nvidia [https://the-decoder.com/meituans-longcat-2-0-shows-china-can-train-massive-ai-models-without-nvidia] * DeepSeek's DSpark boosts AI speed by up to 85 percent [https://the-decoder.com/deepseeks-dspark-boosts-ai-speed-by-up-to-85-percent-a-strategic-win-under-tightening-us-export-controls] * Taiwan raids Super Micro offices in probe over Nvidia chip smuggling to China [https://the-decoder.com/taiwan-raids-super-micro-offices-in-probe-over-nvidia-chip-smuggling-to-china] * Meta secretly tested ChatGPT, Gemini, and Character.AI with thousands of minor-perspective crisis prompts [https://the-decoder.com/meta-secretly-tested-chatgpt-gemini-and-character-ai-with-thousands-of-minor-perspective-crisis-prompts] * US campaigns now run on AI at nearly every step, and Europe is drawing a harder line [https://the-decoder.com/us-campaigns-now-run-on-ai-at-nearly-every-step-and-europe-is-drawing-a-harder-line]