Five things that caught my attention this week in AI tools and open-source models

Short summary

Anthropic's Agent View treats parallel Claude Code sessions like CI pipelines. Zyphra proved AMD can train competitive open-weights models with ZAYA1-8B. ServiceNow embedded Build Agent in Cursor/GitHub Copilot with governance-by-default, while existing productivity metrics haven't caught up to AI-assisted workflows; production systems need reliability buffers for remote integrations.

• •Anthropic's Agent View shifts Claude Code toward multi-task orchestration
• •Zyphra's AMD-trained ZAYA1-8B breaks open-source GPU vendor monoculture
• •ServiceNow embeds governance into IDE integrations rather than leaving it to IT
• •Production workflows need local fallbacks alongside MCP integrations for reliability
• •DORA metrics don't capture AI-assisted code review; new frameworks needed

Generated with AI, which can make mistakes.