Both Claude and ChatGPT prepping major interface updates
🤔 “Should I switch to Claude from ChatGPT?” Share Handy AI with your family and friends to help them understand the crazy world of modern artificial intelligence (and save you some time).
last week’s top stories
🎨 Claude Code “Epitaxy” surfaces as power-user desktop overhaul. Post-April Fools speculation, TestingCatalog says Epitaxy includes a Cowork-style layout with Plan, Tasks, and Diff panels, multi-repo support, and a Coordinator Mode for parallel sub-agents. Anthropic and OpenAI both ship desktop updates next week. Read more
🧠 OpenAI merges ChatGPT, Atlas, and Codex into desktop superapp. TestingCatalog’s April 10-11 reports show OpenAI folding chat, browser, and coding into one app with a Scratchpad UI that fires parallel Codex tasks from a TODO list. Snowflake emojis across employee feeds hint at a “Glacier” model reveal alongside. Read more
🔥 Sam Altman’s home hit with molotov and gunfire in 48 hours. Daniel Moreno-Gama torched a gate at Altman’s Russian Hill residence April 10, then threatened to burn OpenAI HQ. Sunday morning a passenger in a Honda fired rounds at the same property; police arrested Amanda Tom and Muhamad Tarik Hussein. Read more
🕸️ Anthropic withholds Mythos Preview, launches Glasswing cybersecurity coalition instead. Mythos Preview found thousands of zero-day vulnerabilities across every major OS and browser, including a 27-year-old OpenBSD flaw. On April 7 Anthropic handed the model to AWS, Apple, Cisco, CrowdStrike, and 40+ partners rather than releasing it to the public. Read more
✨ Meta debuts Muse Spark, abandons Llama’s open-source playbook. Alexandr Wang’s nine-month rebuild produced a multimodal reasoning model that scored 42.8 on HealthBench Hard and beat GPT-5.4. Contemplating mode fires parallel sub-agents for complex tasks. Meta closed the source code, promising open weights on some future variant. Read more
💎 Anthropic hits $30B ARR, triples compute with Broadcom TPU deal. Run-rate revenue jumped from $9B at end-2025 to $30B, with $1M+ business customers doubling to 1,000 in two months. Broadcom’s April 6 SEC filing commits 3.5 gigawatts of Google TPU capacity from 2027, hedged on Anthropic’s continued commercial success. Read more
💸 OpenAI carves out a $100 ChatGPT Pro tier for Codex. Sam Altman plugged the chasm between $20 Plus and $200 Pro on April 9 with a tier matching Claude Max’s price. Subscribers get 5x Codex usage over Plus, boosted to 10x through May 31. Codex crossed 3 million weekly users. Read more
🏗️ Claude Tasks Mode leak teases Build option aimed at Replit. TestingCatalog screenshots show an Agent Mode toggle replacing Claude’s blank canvas with a five-workflow dashboard: Research, Analyze, Build, Write, Plan. Build competes with Replit and V0 for UI generation. A/B rollout only, no official Anthropic confirmation. Read more
👓 Apple tests four smart glasses designs, targets 2027 launch. Bloomberg’s Mark Gurman reported April 12 that Apple has four frame styles under testing: two rectangular, two oval, in black, ocean blue, or light brown. Acetate bodies, oval camera arrays surrounded by indicator lights, Apple Watch-class silicon inside. Display-less on v1. Read more
💰 Perplexity Computer becomes personal CFO with Plaid banking hookup. April 9 expansion pushed Perplexity’s Plaid integration beyond brokerage to include checking, savings, credit cards, and loans across 12,000 institutions. Stacked on the April 2 IRS tax module, Computer drafts federal returns and builds debt payoff dashboards on request. Read more
🏃 xAI CFO Anthony Armstrong exits amid SpaceX-led shakeup. The Information reported April 9 that Morgan Stanley alum Armstrong departed xAI after six months as part of a broader senior exodus. SpaceX fixers grade employees, Cursor alums Milich and Ginsberg run product engineering. Bret Johnsen keeps combined-entity finance. Read more
🏛️ Appeals court denies Anthropic’s bid to lift Pentagon blacklist. The DC Circuit rejected Anthropic’s stay request April 8, ruling the Pentagon’s authority over vital AI procurement during military conflict outweighs Anthropic’s financial harm. A San Francisco judge had blocked the broader supply-chain risk label last month. Anthropic remains locked out of DOD contracts. Read more
🧪 AI Research of the Week
AI Prediction of Heart Failure Risk from Routine Cardiac CT Scans
University of Oxford
Jake’s Take: There's a little pocket of fat sitting around your heart, and that fat quietly changes texture when your heart muscle is starting to struggle. Years before you'd feel anything. Years before a doctor would catch it. Your eyes can't see the change on a scan.
The Oxford team built an AI that can. They trained it on CT scans from about 59k people in England, then tested it on another 13k or so. It correctly flagged future heart failure about 86% of the time, five years before the condition showed up. People it tagged as high-risk were 20 times more likely to get sick than the low-risk group.
What makes this more impressive is that nobody needs a new test; these are scans hospitals already do for chest pain or other reasons. The AI model just reads the pictures sitting in the system and spots something humans can't. Heart failure usually gets diagnosed late, once the muscle is already damaged. Catching it five years out changes what's possible. The team is now working to make it read any chest CT, not just heart-specific ones.
and then, even more news…
🤖 Anthropic ships Claude Managed Agents, swallows agent infrastructure cost. Claude Managed Agents hit public beta April 8 at $0.08 per session-hour plus standard token rates. The service handles sandboxing, state management, multi-agent coordination, and session tracing. Notion, Asana, Rakuten, and Sentry claim 10x faster time-to-production. Read more