TESS Feed — July 7, 2026

Tue, 07 Jul 2026 20:00:00 -0400

Today was about control of the AI stack: Microsoft reportedly swapped in-house models into apps, China moved toward restricting foreign access to its best models while its labs pushed custom chips, and the strongest technical signal was a new wave of harness, verifier, and agent-runtime infrastructure.

Microsoft starts replacing partner models with its own — A top AI distributor appears to be reducing dependence on OpenAI and Anthropic inside real products, signaling a serious platform shift toward vertically integrated model supply.
China tightens model access as DeepSeek and Zhipu eye custom chips — Chinese AI is looking more sovereign by the day: possible foreign-access restrictions hit global availability at the same time leading labs move to own more of the compute stack.
Harness engineering and verification are becoming core agent architecture — The center of gravity is shifting from just bigger base models to systems that scaffold, check, and iteratively improve model behavior in the loop.
Agent ops stack gets real: managed agents, agent cloud, sandboxes, audit trails — Background tasks, remote MCP, fleet orchestration, secure sandboxes, and tamper-evident runtime logs are turning agents from demos into operable production systems.
Open and local tooling keeps gaining ground against cloud lock-in — From Tencent and DeepSeek model releases to local-first desktop agents, CPU-friendly TTS, and public pushback on vendor dependence, the market keeps rewarding portability and self-hosting.

Full feed (76 items)

TESS Feed — July 6, 2026

Mon, 06 Jul 2026 20:00:00 -0400

Full feed (30 items)