Daily Brief — 31 July 2026
Google puts Gemini Robotics ER 2 into public preview
Google has released two Gemini Robotics ER 2 endpoints in public preview: a general model and an on-device variant. Google says the release adds spatial reasoning, agentic code execution, multi-step tool orchestration, video moment finding, progress classification and multi-robot coordination. The practical change is that robotics teams can now evaluate those capabilities through the Gemini API rather than treating them as a research demonstration — while still accounting for preview-stage volatility. Read Google’s Gemini API release notes.
What else matters
GitHub Models has been switched off
GitHub says its Models playground, model catalogue, inference API and bring-your-own-key option are now retired for all customers, including active users. Developers who used GitHub as a lightweight model-testing surface now need to move their workflows and credentials elsewhere; this is a product shutdown, not a temporary restriction. See GitHub’s retirement notice.
Nscale agrees to buy Anyscale
Nscale has entered a definitive agreement to acquire Anyscale, the company behind the Ray distributed-computing ecosystem. Nscale says the deal adds a software layer for turning compute into an end-to-end AI platform and expands its customer base. If completed, it would join infrastructure capacity with a widely used AI workload-management stack. Read Nscale’s announcement.
Copilot CLI changes its autopilot and sign-in defaults
GitHub’s Copilot CLI v1.0.77 makes unconditional autopilot approval disable the sandbox for the current session when bypass is allowed, and moves browser OAuth to the default login route. The release matters because agent permissions and authentication are operational controls, not cosmetic CLI changes. Read the v1.0.77 release notes.
A new paper finds AI research agents still stumble on the hard parts
A new arXiv preprint reports two case studies of agents conducting open-ended AI research. Its authors find that agents can handle engineering work but struggle with critical parts of the research lifecycle. It is early evidence rather than a production benchmark, but it is a useful warning against equating fast implementation with autonomous scientific judgment. Read the paper and abstract.
