Workstation 博客
AI Agents
标记为 AI Agents 的文章。
From the Hall at GTC Berlin: Huang Says AI Is Now Productive
Workstation engineers were at STATION-Berlin for the 21 October 2026 keynote. Huang's turn was from factories to agents that spend tokens.
Balinder Walia阅读 OpenShell, Announced at GTC Berlin: Policy Outside the Agent
Huang's security film rejected 'please don't' as a control. OpenShell enforces a policy on syscalls, files and network, with Sentry on BlueField-4.
Balinder Walia阅读 
The Berlin Slide That Splits Documents from Databases
Nemotron Parse, Embed, cuVS and Rerank on one row. Auto-Ontology, cuDF and Kumo on the other. Huang's supplier-delay example uses both.
Balinder Walia阅读 
NVIDIA Minerva for AI Factories: Turning Alert Storms into Recovered GPU Hours
One degrading optic can raise thousands of alerts and stall a thousand GPUs. What NVIDIA Minerva does, what it is worth in GPU hours, goodput, MTBI, MTTR and tokens, which of its three consumption models fits, worked examples for a Vera Rubin NVL72 neocloud, a university Blackwell cluster and an enterprise with Grafana and ServiceNow, and how it relates to UFM.
Balinder Walia阅读 
Build AI Agents with NebulaDB: Semantic Search, RAG and a Visa AI Agent
How we built semantic search, RAG and a Thai Visa AI Agent on NebulaDB, with real queries, query plans and retrieval scores measured on our 8.3M-document production server, and the guards that keep private data and invented numbers out of the answers.
Balinder Walia阅读 
TuneBench: How Well Can AI Agents Calibrate Power System Models?
TuneBench tests LLM agents on calibrating generator and battery models within 15 simulations. Only Opus 5.5 and GPT-6 hit median R² above 0.995 in every test. What that figure means, its limits, and how to build the same kind of benchmark.
Agentic AI in banking: efficiency you can defend to a risk committee
Banking AI agents rarely fail on one bad reply; they fail across a session. How to scope tools, test whole conversations against mock systems and measure reliability with pass^k, so the time saved survives audit and production.
Code as infrastructure: one reviewed path for every production change
Infrastructure as code made servers into files. Code as infrastructure goes further: desired state, policy, pipelines and AI agent access all live in Git, are checked by policy, applied by automation and watched for drift. A practical playbook.
Observability for Agentic AI: Tracing, Evaluating and Auditing Agents
AI agents fail quietly: a confident answer from the wrong document, a tool called with the wrong argument. Learn how to trace every model, retrieval and tool step with OpenTelemetry, attach quality scores, and keep the records auditors ask for.
