RuntimeWire

Real-time startup intelligence for the AI economy. Funding, model launches, infra shifts, and founder moves — the signals that matter.

Latest stories — page 21

• Kimi Work secretly attaches raw records from five recent agent sessions to feedback reports

  Moonshot AI’s Windows desktop app packages bounded raw records from the five most recently updated Kimi Work conversations whenever a user submits feedback. The form does not identify those attachments or the conversations selected.

• Deltix pairs AI app exploration with replayable iOS regression tests

  Its Mac agent keeps app binaries local, while screenshots and interface data still travel to Deltix's cloud and Anthropic.

• Owen Van Natta, Facebook's early COO and startup investor, dies at 56

  Dan Rose credited his longtime mentor with negotiating pivotal Amazon and Facebook deals before Van Natta became an early-stage investor.

• OpenAI built a credit bonus into ChatGPT’s experimental desktop onboarding

  A hidden flow asks newcomers to let ChatGPT complete a real desktop task, then can add a remotely configured bonus to their balance.

• DeepSeek open-sources coding-agent harness before V4-Pro API price hike

  Harness v0.1 lets developers swap models, tools and runtimes, while higher V4-Pro rates take effect August 16th.

• California DMV permits Aurora and Kodiak to test autonomous heavy trucks

  Kodiak has begun supervised runs near Mountain View, while Aurora and Kodiak must each complete 500,000 autonomous miles before driverless testing and another 500,000 before deployment.

• Maple said it made DeepSeek V4 Flash available to paid customers

  An undated Maple post announced access for paid customers, while public materials leave the qualifying plans, Proxy availability and model-specific pricing unclear.

• Community attributes animated SVG to Qwen3.8-27B without reproducible run details

  A pelican-on-a-bicycle animation attributed to Alibaba's open-weight model lacks the prompt, raw SVG and execution details needed for independent reproduction.

• GitHub adds xAI's Grok 4.6 to Copilot for long-running coding agents

  xAI's Grok 4.6 is entering GitHub's IDE, command-line and enterprise workflows two days after launch, under usage-based billing.

• Anthropic stole ChatGPT's aura. Then it gave it back.

  Claude Code made Anthropic Silicon Valley's chosen winner. Its models and business remain formidable. Self-inflicted errors and a contested warning about collapsing morale have punctured the sense that the company could do no wrong.

• Joshua Kushner warns AI excitement could weaken venture discipline

  Thrive's first investor letter defends concentrated investing while its deep OpenAI ties put that discipline under a brighter light.

• sandbox.bio ships an embeddable Linux terminal for runnable technical tutorials

  Robert Aboukhalil and Maria Nattestad are turning a five-year bioinformatics teaching tool into infrastructure publishers can add to their own sites.

• 224 Ventures outlines seed strategy for robotics, work software and AI infrastructure

  Shaun Johnson says the $100M-plus firm will seek nonconsensus seed investments with Oriol Vinyals and Yann LeCun.

• Proxylity opened WireGuard endpoints to unknown devices on AWS

  Founder Lee Harding paired open peer enrollment with async Lambda delivery, turning encrypted UDP packets into triggers for long-running workflows.

• Ukraine finds Nvidia Jetson module in Russia's S-71M cruise missile

  Ukraine says the robotics computer may support onboard vision, raising questions about resale routes that move commercial silicon into Russian weapons.

• MiniMax will present H3 at ModCon in San Francisco on August 18th

  Morgan Suo will discuss MiniMax's multimodal video model at Modular's San Francisco conference for AI infrastructure developers.

• Starlink ends Global Roam, pushing ocean sailors toward pricier metered plans

  Sailing La Vagabonde is asking SpaceX to preserve an affordable tier before existing Global Roam service ends on August 17th.

• Marco Arment launches Unforgetful, a persistent-reminder layer for Apple Reminders

  The Instapaper and Overcast creator built a $19.99-a-year Apple Reminders client around persistent alerts and forgiving snooze controls.

• Grok 4.6 tops BBEH Mini benchmark, scoring 0.67 on 460 tasks

  In a 460-task BBEH Mini evaluation, SpaceXAI’s Grok 4.6 led the field with a 0.67 score at $0.0034 per task. Anthropic’s Claude Opus 4.8 followed at 0.62, while OpenAI’s GPT-5.6 Sol Pro posted 0.59.

• Meta releases local Muse Glimmer as Zuckerberg argues for AI 'for everyone'

  Meta released the 30B open-weight model alongside Zuckerberg's case for widely distributed personal AI, while keeping Muse Spark 1.2 behind its products and API for now.

• Cloudflare uses MCP headers to detect and block unapproved agent connections

  Cloudflare Gateway uses MCP protocol headers to identify inspected requests and block connections outside approved routes, while specialist vendors pursue controls closer to individual agents and tools.

• Andor Health adds remote monitoring to Epic Toolbox for CMS ACCESS

  Raj Toleti is packaging devices, AI triage and clinical labor inside Epic as Medicare tests outcome-linked payments through its new ACCESS model.

• Alibaba schedules Qwen3.8-27B open-weight release for August 14th

  Alibaba's Qwen team set August 14th for a 27B dense vision-language model, but the supplied materials do not confirm when downloadable weights will appear.

• Pony.ai and Uber plan more than 2,000 robotaxis across four European cities

  Pony.ai will supply the autonomy stack, Uber will integrate rides into its network, and local partners may own and operate fleets in four cities that remain unnamed.