1. Codex can now drive your Mac, browse the web, and remember what you did last week OpenAI rebuilt Codex this week into a desktop application that operates the user's computer, opens its own browser, generates images, retains memory across sessions, and accepts third-party plugins.

2. A 35B model on a laptop drew a better pelican than Opus 4.7 this week Simon Willison ran his pelican-on-a-bicycle SVG benchmark against Qwen3.6-35B-A3B on his laptop and Anthropic's newly released Claude Opus 4.7 in the cloud. The local model won.

3. €54,000 of Gemini calls in 13 hours, charged to a developer who never made them A developer posted to Google's AI forum that their Firebase project accrued roughly €54,000 in Gemini API charges over 13 hours. Attackers had found the browser-side API key embedded in their web app.

In Brief

• OpenAI launches GPT-Rosalind for life sciences research OpenAI released a frontier reasoning model aimed at drug discovery, genomics analysis, and protein reasoning workflows. It targets research labs rather than general developers.

• Physical Intelligence ships π0.7 robot model that handles untrained tasks The startup released a robot control model it says generalizes to tasks it was not explicitly trained on. Physical Intelligence frames it as an early step toward a general-purpose robot brain.

• AI referral traffic to US retailers jumped 393% in Q1 Adobe data shows AI-driven visitors to retail sites rose 269% in March alone, and they convert at higher rates than standard traffic. Retailers are now optimizing directly for chatbot referrals.

• Anthropic CPO Mike Krieger exits Figma board before launching competing product Krieger stepped off Figma's board after reports he plans a rival design tool. The move fuels investor fears that top AI labs will swallow vertical SaaS categories.

• Factory raises $150M at $1.5B valuation for enterprise AI coding Khosla Ventures led the three-year-old startup's round as it pushes coding agents into large companies. Factory positions against Cursor, Cognition, and Codex in the enterprise tier.

• Upscale AI reportedly in talks to raise at $2B valuation The AI infrastructure company is pursuing its third round in seven months since launch. Investor demand for inference and deployment layers continues despite rising model costs.

• InsightFinder raises $15M for AI agent failure diagnosis The startup builds observability tools that trace where agents break down across full production stacks. CEO Helen Gu says diagnosing failures now requires visibility beyond the model itself.

• Canva AI assistant gains tool-calling for editable designs The update lets users generate editable Canva designs from text prompts by invoking the platform's existing tools. It pushes Canva further toward agent-built workflows instead of template selection.

• Roblox adds agentic planning and testing to its AI assistant The new tools let creators plan, build, and test games end-to-end inside Roblox Studio. Roblox is pitching it as a full-pipeline assistant rather than a code helper.

• DeepL adds real-time voice translation for meetings DeepL extended its translation stack to live speech, targeting Zoom and Microsoft Teams integrations. The feature puts it in direct competition with Google and Microsoft's built-in live translation.

• Runway CEO pitches 50 AI films for the cost of one blockbuster Cristóbal Valenzuela argued studios should use generative video to produce dozens of mid-budget films rather than a single $100M release. He framed volume as a hedge against hit-rate risk.

Read the full edition →