Google Open-Sources Harness² AI Agent Framework

Key Takeaways

• Google open-sourced Harness², an AI agent framework with 8 editable components
• A single refinement step improved performance across all tested benchmarks
• Voxel51 3.0 offers a free tier with 3 datasets of 10,000 samples each
• Wisedocs improved its MLCR benchmark score from 20% to 75%
• Tech stocks now make up about 44% of total S&P 500 value
• Nvidia is the most valuable company at roughly $7.5 trillion
• Apple gained over $1.4 trillion in six months, growing 30%
• Rivian is expanding into robotics and licensing its tech to other brands
• 70% of US consumers are concerned about AI-assisted pricing at retailers
• 71% of economists believe AI will significantly boost productivity

Google Open-Sources Harness2 Recursive Agent Framework

Google Cloud AI Research released Harness², an open-source framework that improves AI agents without training new models. It includes 8 editable components like system prompts, memory, and plugins. Version 0.1.0 runs on Linux with Python 3.12 or 3.13. A single refinement step improved performance across all tested benchmarks.

Voxel51 3.0 Launches Multimodal Data Platform for Physical AI

Voxel51 released version 3.0, an end-to-end platform for search, curation, annotation and evaluation of multimodal data. It includes Voxel51 Agent with 45+ MCP tools, 80+ operators and 18 open-source skills. A free tier offers 3 datasets with 10,000 samples each. The platform supports teams training models on robot, vehicle and factory sensor data.

Wisedocs Says Insurance Claims Still Challenge AI

Wisedocs reported that fragmented, fax-heavy insurance claim files remain a major challenge for AI. Its MLCR benchmark score improved from 20% to 75% over several months. A single PDF claim can reach 8 gigabytes and duplicate records can exceed 50%. The company cited over $7 billion spent globally on fax systems.

Si.gov Pushes Super Intelligence Without Detailed Costs

Si.gov rebranded federal AI efforts as Super Intelligence around the Genesis Mission. The initiative unites 15 federal agencies with industry, academia and Department of Energy labs. The site did not publish budgets, metrics or security models. No capability was demonstrated and no specific systems or versions were listed.

AI Stock Boom Creates Lopsided US Economy

Experts warn that AI's dominance in the US stock market creates economic risks. Tech stocks now account for about 44 percent of total S&P 500 value. Nvidia is the most valuable company at roughly $7.5 trillion. Apple grew 30 percent, gaining over $1.4 trillion in six months. The combined net worth of major AI executives exceeds $2.5 trillion.

Apollo Cuts CLO Portfolio Work From Days to Minutes

Apollo's Redding Ridge Capital used Rowspace agents to speed up CLO portfolio construction. The work went from two to three days in Excel down to minutes. Mock portfolios of 200 to 400 names now reach 80 to 90 percent completion before portfolio managers step in. Weekly trade reconciliation that once took hours for two to three people now takes about one minute. Human QA remains part of the daily review process.

US Consumers Worry About AI Pricing Tools at Retailers

US consumers are growing upset about retailers and restaurant chains using AI-assisted pricing tools. A survey found 70 percent of consumers are concerned about AI being used to set prices. About 60 percent believe the tools are unfair. Around 55 percent would pay more if prices felt transparent and fair. The article calls for retailers to be open about how they use AI in pricing.

Rivian Aims to Become a Broader Tech Supplier

Rivian CEO RJ Scaringe wants the company to grow beyond making electric vehicles. Rivian has built its own AI models and is expanding into areas like industrial robotics through Mind Robotics, which is 31 percent owned by Rivian. The company is also working on robotaxis with Uber and exploring licensing deals with VW. Scaringe sees Rivian becoming a tech supplier that licenses its electric architecture, software, and autonomy stack to other brands.

AI Super PACs Focus on Building Political Friendships

Major AI campaign spending groups avoided confrontational races this election cycle. Instead, the two big AI super PAC networks focused on building relationships with lawmakers. They were launched last year with plans to follow a strategy similar to the crypto industry's political playbook. That approach originally included targeting opponents in high-profile races. The article describes how these groups shifted toward a friendlier approach in politics.

Economists See AI Driving Global Economic Growth

A survey of 94 economists by the FT found that 71 percent believe AI will significantly boost productivity. About 45 percent expect AI to be a major driver of economic growth. The benefits are expected to spread beyond the US and China to industries like manufacturing, services, and finance. However, 60 percent of respondents also expect AI to increase inequality. The survey was conducted with Oxford's Saïd Business School and the London School of Economics.

Sources

• Google Open-Sources Harness2, a Recursive Agent Harnessing Framework
• Meet Voxel51 3.0: A Unified Multimodal Data Platform for Physical AI
• Wisedocs says insurance claims still break AI
• Si.gov Pushes Super Intelligence, Hides Hard Costs
• Soaring AI stocks create lopsided US economy
• Apollo Turned Three Days Into Minutes With Agents
• Consumer apprehension grows as US retailers adopt AI-assisted pricing tools
• TechCrunch Mobility: Rivian's next act
• AI, meet your freshman class
• Global economy set for AI-driven surge, FT survey suggests