Monday, November 16, 2026 Edition No. 240
Search
Daily News Fully automated · 5 min read

Google AI Updates: November 14, 2026

Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled

1. Google Research Details /teamwork in Antigravity, a Multi-Agent System Behind Machine-Checked Proofs and Large Rust Rewrites

Google Research published results on November 13 from /teamwork, a multi-agent system in Antigravity applied to open problems in mathematics, chemistry, systems software and ML engineering with human researchers framing and verifying the work. Parallel agents explore candidate solutions while dedicated falsifiers check them, including the Lean 4 proof checker, unmodified test suites, simulators and hardware counters, and critics review results from clean contexts without the producer’s reasoning. Workflows are declarative natural-language patterns such as Long Proof, Deep Review, /boost and DuoDiscover, running on Gemini 4 Argon, Carbon and Hydrogen. Reported results include a full proof of the 2013 Courtade-Kumar conjecture formalized in Lean 4, a machine-checked formalization of the Weil conjectures of about 1.8 million lines of Lean across 28,800 subagent invocations, about 800,000 lines of safe Rust added to the Zircon kernel in two weeks that pass all kernel unit tests and boot on arm64, and a Rust port of the libgav1 AV1 decoder that ended 20% to 30% faster than the C++ original. Inside Google, two shipped Borg scheduling changes cut TPU workload preemption by 24% on some hardware types. Gemini 4 models with /teamwork and /boost, plus Deep Review, are available in Antigravity now, and a program offers 10,000 academic researchers six months of free access to the highest usage tier; the formal proof and rewrite patterns are not yet released externally. Source

2. Google Cloud Brings Agent Telemetry for BigQuery, Looker, ADK and LangGraph Agents Into Customer-Owned BigQuery Tables

Google Cloud announced expanded agent observability and evaluation across its Data Cloud on November 13. First-party Conversational Analytics agents emit telemetry with no setup, and queries routed to BigQuery data agents in Gemini Enterprise through A2A workflows, now GA, send Cloud Monitoring metrics, Cloud Trace spans and conversation logs to the agent’s project. Conversation logging in BigQuery is GA: when an agent owner opts in, prompts, responses and turn-level telemetry stream into a customer-owned agent_events table that inherits the dataset’s IAM, CMEK and VPC Service Controls, and logging is skipped for any turn whose knowledge source uses row-level security, column-level policy tags or dynamic data masking. Custom agents built on ADK or LangGraph can stream full tool-invocation trees into BigQuery through the open-source BigQueryAgentAnalyticsPlugin. The Conversational Analytics API is now GA, as are agent monitoring dashboards in BigQuery Studio and Looker System Activity dashboards that track engagement and token usage. For cost control, agents accept a per-query byte scan cap (big_query_max_billed_bytes), and daily input and output token quotas per project, user or agent are in preview. The post also covers Prism, an open-source tool for benchmarking Conversational Analytics agents with reusable test suites. Source

3. Google Cloud Pairs Cluster Toolkit With Agent Skills for HPC and Offers AlphaEvolve on Cloud Batch Ahead of SC26

Google Cloud outlined its HPC and AI lineup on November 13 ahead of the SC26 conference (November 15 to 20, Chicago). Cluster Toolkit is now paired with composable agent skills, downloadable from the Google Skills and DeepMind Science Skills repositories, that direct an agent to audit compute and storage quotas, run canary checks for regional capacity, match the workload to a validated blueprint and wait for human approval before running gcluster deploy. Domain skills cover OpenFOAM and Siemens STAR-CCM+ fluid dynamics runs, GROMACS and AlphaFold 3 pipelines, and Devito and Open Porous Media subsurface simulations, with agents operating inside IAM, Kueue quota and budget guardrails. An AlphaEvolve on Cloud Batch solution provisions the AlphaEvolve controller from a single blueprint and fans out hundreds of candidate evaluations concurrently across single-VM sandboxes or multi-node MPI task groups. On infrastructure, the post lists A5X bare metal instances on NVIDIA Vera Rubin NVL72 as newly available, eighth-generation TPUs (TPU 8t for large-scale pre-training and TPU 8i for inference), Batch support for single-host TPU slices, and Managed Lustre with up to 10 TB/s of read throughput. Source