Kraken's Engineering Operations Lead Nik Sudan discusses how to move agentic AI projects from pilot to production without compromising code health. Key topics include why raw AI adoption metrics are vanity metrics, using tools like the LinearB MCP server to combine engineering metrics with repository data to find workflow bottlenecks, and how to translate P90 cycle times into business narratives for non-technical stakeholders. The newsletter also covers the 'discernment horizon' concept from Steve Yegge, loop-driven development from Anthropic's whitepaper, why model divergence reveals prompt ambiguity rather than tool quality, and Google's Open Knowledge Format for structuring agent context.
Table of contents
1. Are we approaching the discernment horizon?2. A magical allegory for tech disruption3. Evolving beyond the CI/CD pipeline4. Why model divergence is actually a feature5. Life beyond tokenmaxxing6. Structuring knowledge with plain text7. Midjourney’s leap into medical imaging101 Impressions