Signal from the noise. Curated for Anton.
A packaging error in Claude Code v2.1.88 shipped unobfuscated source maps containing ~512k lines of TypeScript. The internet mirrored it instantly — Anthropic has issued 8,000+ DMCA takedowns but can't contain the spread. Devs have found unreleased features including "Coordinator Mode" for multi-agent orchestration, an "Ultra" subscription tier, and most controversially: an inactive "Undercover mode" that allegedly let Anthropic employees contribute to open-source repos as AI agents without disclosure. "More open than OpenAI" — except this time they didn't mean to be.
AI LEAKTwo moves that reshape the landscape. Google dropped Gemma 4 under Apache 2.0 on April 2 — four model sizes, full Gemini-era architecture, most permissive license yet. Meanwhile Microsoft launched three MAI models (speech-to-text, voice gen, image) and confirmed plans to build its own frontier models by 2027. Translation: the OpenAI dependency is officially a liability. Copilot adoption sits at a dismal ~2% among Office users, and their best new product (Copilot Researcher) chains GPT + Claude together — the irony writes itself.
AI INDUSTRYOne year after Trump's tariff blitz, the scorecard is ugly. SCOTUS struck down the IEEPA tariffs in February (6-3), ~$166B in duties being refunded. Despite this, Trump marked the anniversary with new 100% pharma tariffs. The data: 100k manufacturing jobs lost since Jan 2025, consumer confidence cratered, trade deficit barely moved. Supply chains simply rerouted through Vietnam. The "TACO" label (Tariff Anxiety Creating Outflows) has stuck on Wall Street. The contrarian take gaining traction: the tariffs did reshape trade — just not toward the US.
GEOPOLITICS TRADETrump's DOJ is appealing the ruling that blocked the Pentagon's retaliation against Anthropic for refusing autonomous weapons contracts. An extraordinary coalition — Microsoft, retired military leaders, EFF, Catholic theologians — filed friend-of-court briefs supporting Anthropic. The contrarian take: Anthropic's principled stand is also brilliant commercial strategy — enterprise trust skyrocketed because they said no to DoD.
Court docs from Musk's OpenAI lawsuit reveal Feb 2025 texts where Zuckerberg offered to help DOGE — specifically offering Meta teams to remove content threatening DOGE personnel. Musk responded with a heart emoji. Zuckerberg also discussed (then declined) joining Musk's $97.4B OpenAI bid. The cage match was always theater.
Russia launched its largest single-day strike in months — 286 drones and missiles on April 3 — after the Kremlin rejected Zelenskyy's proposed Easter ceasefire. Ukraine intercepted ~260. Meanwhile the EU approved a €90B loan package, and European defense spending rose 20% for the second consecutive year. Europe is rearmament-mode, permanently.
Cursor shipped v3 on April 3 — a ground-up redesign centered on AI agents that plan and execute across entire projects. Multi-repo workspaces, task delegation, moves beyond autocomplete into agent-managed workflows. The IDE wars just became the agent wars.
OpenAI's DoD deal aftershocks continue — robotics chief Caitlin Kalinowski resigned saying they "moved too fast on something too important." The app bounced back to the top of charts, proving convenience wins eventually. But the Anthropic contrast is now the central moral axis of the industry.
First full-length in six years, out April 3 via Brainfeeder. Produced primarily by Greg Kurstin. Dropped alongside new Arlo Parks and Dermot Kennedy — stacked New Music Friday. Thundercat's bass-drenched cosmic funk remains unmatched.
Firefly now lets creators train custom models on their own artwork to replicate their personal style — a genuine game-changer for designers needing brand consistency. Creator-controlled AI tools vs. open-source generators: Adobe is betting on the trust axis.