RuntimeWire

Featured

• Specific Labs launches Real-SWE, where 38.8% is enough for first place

  Specific Labs co-founders Janak Sunil and Siddhant Paliwal built the benchmark from licensed production code, turning a recurring lab question into a data product.

• Attackers claim they are publishing Revolut customer data to force a payout

  Screenshots show identity documents and verification photos; Revolut says its systems and customer funds were unaffected.

• OpenAI wants a head-to-head with Claude Code—using your own work

  Build 8881 targets successful Claude Code importers with a plugin that reruns an imported task in Codex for a direct comparison.

• Mercor spends 3X employee salaries on LLM inference, CEO says

  Brendan Foody says the compute is adding headcount, while Mercor separately says it pays contractors more than $4M a day.

• Higgsfield puts GPT-driven motion graphics inside After Effects

  Alex Mashrabov's AI Motion Designer turns chat instructions into editable motion graphics in an open Adobe composition and requires Higgsfield, ChatGPT and Adobe subscriptions.

• Sunday Letter from the Editor: RuntimeWire moves to mobile, leaving its Google search bet behind

  Original reporting slowed this week as I prepared the apps for launch. The plan now is native mobile, real-time alerts and more investigations, not waiting for search traffic to return.

• Positron raises $875M to get its next inference chip into silicon

  Thomas Sohmers' three-year-old chipmaker is valued at $5B, with Asimov tapeout due by year-end and production targeted for the second half of 2027.

• Startup Spotlight: Orchestra turns production traces into smaller models; 100x remains a claim

  Founders Luis Manrique and Aamir Poonawalla want repetitive AI work to produce owned models instead of permanent frontier-model bills. First-party benchmarks show narrower gains, while named customers and pricing remain undisclosed; Dealroom reports a $125,000 YC investment.

Investigations

• OpenAI wants a head-to-head with Claude Code—using your own work

  Build 8881 targets successful Claude Code importers with a plugin that reruns an imported task in Codex for a direct comparison.

• Kimi quietly sends full user prompts into ByteDance’s China-based analytics pipeline

  I intercepted a verbatim Kimi Work 3.2.7 prompt inside a `kimi_work_send_message` event uploaded to Volcano Engine's DataFinder analytics system, alongside user, conversation and device identifiers. The company's specific disclosure for the analytics SDK lists device, network, location and application information, without specifically naming conversation text.

• Microsoft let Anthropic move in while OpenAI was sleeping on the couch

  An unreleased Claude for Windows build shows Anthropic wiring itself deeper into Microsoft’s enterprise identity stack, as the partnership that defined the first years of the AI boom becomes increasingly less exclusive.

• Microsoft is falling asleep at the wheel as hidden code reveals Google’s big Windows AI push

  Google once went to antitrust regulators to pry open Microsoft’s desktop search. Nearly 20 years later, hidden code shows it building an AI layer across Windows itself.

• OpenAI is preparing to pay you in ChatGPT credits for using Stripe, HubSpot and other plugins

  A gated Windows flow requires users to install a selected plugin and complete a successful action before credits reach their workspace.

• Antigravity code reveals Gemini-powered Vetted mode

  Version 2.12.2 contains a gated permission preset that assigns a second AI system to approve or reject an agent's proposed actions.

View all investigations

Stories

• Jacob Coxon quits Anthropic before equity vests over AI risk

  The 27-year-old pretraining researcher worked at OpenAI and Anthropic, then publicly accused both labs of racing toward self-improving systems.

  0 upvotes

• StemJSON lets AI add native modules after mobile apps ship

  Vasyl Krychun's open JSON language renders generated modules through SwiftUI and Jetpack Compose, bringing app-store review and runtime trust into the prompt-driven development pitch.

  0 upvotes

• Yardi buys Sidero Labs as Talos Linux adds a native hypervisor

  Sidero president Steve Francis says the acquisition will support Talos Linux's move into virtual machines and standalone containers; an alpha release is due at TalosCon in October, with general availability planned for December.

  0 upvotes

• Buildots raises $130M to turn construction cameras into a mega-project control layer

  Roy Danon and his Talpiot co-founders are widening Buildots from visual progress tracking into workforce and field operations for data centers, fabs and hospitals.

  0 upvotes

• Archsloth fixes two AutoRound flags, says Qwen3-4B drifts less

  The project reports 20.9% to 54.4% lower KL divergence at the same 2.50 GB size, with calibration text doing more work than bit allocation.

  0 upvotes

• Arena ranks GPT-6 Astra Max first for web development, Claude Fable 5.1 Max first for agents

  Arena's founders are turning model evaluation into a business by dividing "best" into tasks, costs and confidence intervals.

  4 upvotes

• Highstock raises $30M to take its surplus marketplace into apparel

  a16z is leading the Series A as Camille van Horne expands Highstock beyond beauty and tests its inventory-matching software in a larger category.

  4 upvotes

• Attackers claim they are publishing Revolut customer data to force a payout

  Screenshots show identity documents and verification photos; Revolut says its systems and customer funds were unaffected.

  5 upvotes

• Chift raises €10.5M to wire Europe's fragmented financial software together

  BlackFin led the round as Chift expands beyond 120 connected financial systems and develops integrations for AI products that need financial data.

  0 upvotes

• Agentic AI Foundation launches MCPA certification for MCP builders

  David Soria Parra's open integration protocol now has an exam covering architecture, execution, permissions, and trust boundaries.

  0 upvotes

• Mercor spends 3X employee salaries on LLM inference, CEO says

  Brendan Foody says the compute is adding headcount, while Mercor separately says it pays contractors more than $4M a day.

  2 upvotes

• Sunday Letter from the Editor: RuntimeWire moves to mobile, leaving its Google search bet behind

  Original reporting slowed this week as I prepared the apps for launch. The plan now is native mobile, real-time alerts and more investigations, not waiting for search traffic to return.

  1 upvote

• Startup Spotlight: Orchestra turns production traces into smaller models; 100x remains a claim

  Founders Luis Manrique and Aamir Poonawalla want repetitive AI work to produce owned models instead of permanent frontier-model bills. First-party benchmarks show narrower gains, while named customers and pricing remain undisclosed; Dealroom reports a $125,000 YC investment.

  1 upvote

• Positron raises $875M to get its next inference chip into silicon

  Thomas Sohmers' three-year-old chipmaker is valued at $5B, with Asimov tapeout due by year-end and production targeted for the second half of 2027.

  1 upvote

View all Stories