AI Agenda: New Agents and Investments

Defense Industry and AI Policies

This week, relations between AI companies and the U.S. Department of Defense (Pentagon) developed in different directions. While Anthropic sued the Pentagon, objecting to being defined as a supply chain risk, Microsoft submitted a legal opinion supporting Anthropic in this process. On the other hand, Google announced that it will provide Gemini-powered AI agents (Agent Designer) so Pentagon employees can automate repetitive administrative tasks.

New Tools, Features, and Models

The data processing and presentation capabilities of large language models are expanding. Both OpenAI’s ChatGPT and Anthropic’s Claude model can now create interactive charts and diagrams to explain complex scientific and mathematical data.

On the model side, multimodality and processing speed stand out. Google released Gemini Embedding 2 in preview, which can search text, images, video, and audio in a single system. While Hume AI introduced the on-device TADA model, which reduces hallucinations by synchronizing text and audio; NVIDIA launched the open-source 120B Nemotron 3 Super reasoning model with a 1 million token context window for multi-agent workflows.

Advancing AI Agents and Systems

The following were the notable developments in automation and assistant systems (AI Agents):

• Anthropic Code Review: The multi-agent code review system integrated into Claude Code focuses on finding logical errors in pull requests that escape the human eye.
• Perplexity Personal Computer: A hardware-software integration similar to OpenClaw was introduced, combining Perplexity with local apps through an always-on Mac mini.
• Nvidia NemoClaw: Launch preparations began for the open platform where enterprises will be able to run AI agents on any hardware infrastructure.
• Cloudflare /crawl API: The company known for anti-bot services released a new API endpoint that allows websites to be crawled with a single call.

Investments, Infrastructure, and Acquisitions

Investments in AI compute infrastructure and agent-based platforms continue at full speed across the sector. Thinking Machines Labs, founded by former OpenAI executive Mira Murati, signed a gigawatt-scale infrastructure agreement with Nvidia to meet hardware needs. On the funding side, AI infrastructure provider Nscale received $2 billion, software development platform Replit $400 million, and legal-focused automation startup Legora $550 million.

The security and ecosystem expansion strategies of large companies are also becoming clear. OpenAI acquired security startup Promptfoo to protect AI agents against cyber threats; Meta acquired social network platform Moltbook, where AI agents interact with each other.