1. DeepSeek Updates V4.1 Flash to Shorten Task Execution and Output, Releases Production-Grade Protocol Toolkit DeepSeek has released DeepSeek-V4.1-Flash-1017, a minor update to its V4.1 Flash model that the company says improves reasoning, coding, agentic tasks and visual benchmarks.

2. Claude Code team shifts most code reading to machines as merges more than double and revert share falls over 40% Anthropic says Claude writes about 90% of Claude Code’s own code, creating more changes than its engineers can practically review line by line.

3. FDE-Gym Tests AI Delivery on Real-World Customer Projects: Best Configuration Solves Only 14 of 95 Tasks A new benchmark called FDE-Gym tests whether AI agents can complete the work of forward-deployed engineers—specialists who turn customers’ often vague requests into functioning systems.

In Brief

• Replit Makes OpenAI Its Preferred AI Partner AI coding platform Replit made OpenAI the default provider for Replit Agent, citing performance, cost, and partnership strength, while continuing to offer Anthropic models as options.

• Meta Releases Open-Weight Muse Spark Open for High-Memory Macs Meta launched Muse Spark Open, an 816-billion-parameter multimodal model designed for coding and agentic tasks, with a one-million-token context window and local Mac support requiring at least 1TB of RAM.

• Meta Removes Disputed Web-Scraping Language From Its Commercial Terms Meta deleted language that legal experts and publishers interpreted as granting it permission to scrape websites for AI training. The company said the terms applied only to ads and business products and did not override publishers’ robots.txt preferences.

• Amazon Launches Bee Wearables for Ambient AI Assistance Amazon introduced Bee devices that passively process conversations to extract notes, reminders, and tasks, then delete the raw audio; the resulting transcripts are encrypted with user-controlled keys. Devices chief Panos Panay said Amazon expects ambient AI to handle more tasks without requiring users to open individual apps.

• Rivian Plans to Bring Third-Party AI Agents Into Its Vehicles Electric-vehicle maker Rivian plans to connect agents from outside AI companies to its in-car assistant, although CEO RJ Scaringe gave no launch date. Rivian says agents will not control safety-critical functions such as braking or steering.

• Major Websites Block AI Agents From Transactions and Accounts Yelp, Delta Air Lines, JPMorgan Chase, and Charles Schwab confirmed restrictions on AI agents, citing licensing requirements or security concerns, while Amazon continues litigating against agent-based shopping. Meta has instead secured agent access through partnerships with retailers including Walmart, Gap, Sephora, and Best Buy.

• Anthropic Reports Fewer Junior Technical Hires and More Meeting Time Anthropic said junior employees’ share of its technical workforce fell from 21% to 14% since early 2025, while junior candidates’ share of technical hiring dropped from 27% to 9%. Its report also found meeting time per employee rose 20% year over year as staff spent more effort reviewing AI output and coordinating decisions.

• Jev Reaches 35% of Vercel’s Paid Teams Within Two Weeks Vercel said Jev, a lightweight TypeSafe AI model that selects among predetermined outputs, became its fastest-adopted model, reaching 35% of paid teams within two weeks. The figures cover traffic through Vercel’s AI Gateway and may undercount models accessed directly from their developers.

• Elorian Releases Glimpse Visual-Reasoning Model AI startup Elorian launched Glimpse as a research preview with API and chat access; the model can revisit, crop, and measure images or retrieve video frames while reasoning. Elorian says Glimpse targets robotics, scientific, media, and creative organizations with substantial visual data.

• Nvidia’s RTX Spark Laptops Bring Large Local Models to Windows on Arm The first laptops using Nvidia’s 20-core RTX Spark N1X processor offer up to 128GB of unified memory for local AI workloads, with tested prices ranging from $3,199.99 to $6,999.99. Early testing found broad everyday-app compatibility but remaining inconsistencies in gaming and AI workloads.

• MJev-Spatial Uses Single-Pass Classification for Faster Visual Decisions Researchers introduced MJev-Spatial, a 4-billion-parameter model that handles spatial choices, numerical estimates, and image-pointing tasks without generating answer tokens. They report that it led seven of 12 benchmarks and ran 5.6 times faster than its autoregressive Qwen3.5-4B base model.

Read the full edition →