Nvidia introduces Nemotron 3 Nano Omni with vision and speech for powerful agentic AI use. Nvidia releases Nemotron 3 Nano Omni, a 30B Mixture-of-Experts model that unifies text, vision, and speech for low-latency agentic AI. Outcome engineers get a compact multimodal frontier model to run perception+action agents at lower inference cost — important for building legible landscapes and agent orchestration (Principles 06, 09).

Laguna XS.2 and M.1: A Deeper Dive. Poolside publishes Laguna M.1 and an open-weight Laguna XS.2 plus a runtime focused on long-horizon, code-writing agents. This gives practitioners modifiable MoE models and runtimes for agentic coding pipelines, letting teams iterate on artifact-producing agents and validate outputs locally (Principles 03, 07, 16).

With agents on the rise, is the ‘modern’ data stack already legacy infrastructure?. Google Cloud launches Agentic Data Cloud to rearchitect the data stack around agents as first-class consumers. Outcome engineers must redesign data contracts, retrieval latency, and observability so agents can reliably access, reason over, and audit grounding sources in production (Principles 06, 09).

Interview — Sam Altman and Matt Garman on Bedrock Managed Agents. AWS and OpenAI outline Bedrock Managed Agents as a route to speed enterprise agent deployments and support multi-cloud strategies. If you adopt managed agents, plan for orchestration, provenance capture, and artifact graphs — agentic coordination becomes an organizational concern, not just an engineering one (Principles 09, 11).

Warp is now Open-Source. Warp open-sources its agentic IDE so developers can run built-in or custom CLI agents and extend terminal-first workflows. Teams can integrate these local agents into CI, reproducible artifact flows, and developer pipelines, lowering friction to ship and validate agent-driven features (Principles 03, 09).