The frontier is being built to run — Aug 25

Five Aug. 24 signals show frontier AI being built for control and execution: owned domain models, open weights, checked developer workflows, agentic silicon, and physical embodiment.

Thomson Reuters is taking model ownership into professional software. Thomson is a $40 million in-house LLM trained on a strong open-source foundation, proprietary legal and tax content, and subject-matter expertise. Thomson Reuters plans to deploy it first in Tabular Analysis inside CoCounsel Legal, while a small version is available on Hugging Face for academic and non-commercial use.1

Poolside is making the open-weight argument operational. Laguna S 2.1 is a 118-billion-parameter coding model for agentic work. Poolside says the model matches or exceeds several larger models on Terminal-Bench 2.1 and SWE-Bench Pro, runs on one NVIDIA DGX Spark, and is available through Hugging Face, its API, and OpenRouter.2

OpenAI is selling GPT-5.6 inside a development process. OpenAI says GPT-5.6 Sol, Terra, and Luna are now available in Kiro, where requirements, codebase context, review checkpoints, and property-based testing surround the coding agent. OpenAI reports roughly 82% lower cost for successful GPT-5.6 Terra tasks on Terminal-Bench 2.1 in Kiro.3

Intel is designing hardware around concurrent agents. Intel's Aug. 24 architecture release describes Diamond Rapids, Crescent Island, and Wildcat Lake. The designs emphasize cores, memory, inference engines, chiplet interconnects, and edge NPUs for workloads with larger models, longer context windows, and more agents running at once.4

Physical AI is attracting a sharper valuation bet. TechCrunch reports that General Intuition is in talks to raise at a $6 billion pre-money valuation, weeks after a $320 million round at a $2.3 billion valuation. The round was still being finalized, and the company plans to use the funds for its general model, robotic embodiments, compute infrastructure, and hiring.5