Field Notes

A detailed architecture for a GTM system where Codex reads weekly outcome logs, proposes single scoring or prompt changes backed by evidence, validates them against an eval gate, and opens a PR for human review — keeping sending and merging strictly outside the automated loop.

A compact framework for reliable agent orchestration built around five layers (prompt, context, harness, workflow, backpressure), with concrete primitives for reversibility, token budgets, vertical task decomposition, and protecting an orchestrator's own context.

Sierra built Pinecone, an internal cloud agent platform that unifies employee AI usage into a single shared, durable, multiplayer system rather than individual laptop-based agent sessions. The design philosophy centers on building durable primitives (context, environments, tools) rather than workflows, centralizing improvements company-wide, and treating accumulated workforce knowledge as the key competitive moat as frontier model intelligence commoditizes.

Sierra's engineering leaders detail how they built a single MCP gateway connecting AI agents to internal tools (Slack, GitHub, Salesforce, etc.), covering coordination strategy, agent verification pitfalls, cross-customer data safety, identity models, and adoption results (89% of employees, 45 services, two-thirds of commits now from other teams).

Cathryn Lavery argues the highest-leverage first AI loop isn't a more autonomous agent — it's a system that reads your own Claude Code/Codex session transcripts as evidence, surfacing repeated corrections, tool failures, and workflows worth turning into content, config fixes, skills, hooks, or slash commands. She open-sourced 'agent-improvement-loop,' a local tool that scans sessions, stages proposals into seven categories, and requires manual approval before any change — pairing with a companion piece on why agent-native CLIs beat official APIs/dashboards for a second, non-human user class.