Kyle Daigle

AI Summary

→ WHAT IT COVERS GitHub COO/CMO Kyle Daigle covers GitHub's scaling crisis (commits growing from 1B to 14B projected annually), the evolution of Copilot toward ambient agentic workflows, internal AI productivity systems using MCP servers and micro-skills, open source trust mechanisms, NPM security tradeoffs, and Microsoft's developer platform strategy around Build 2025. → KEY INSIGHTS - **Micro-skills over mega-skills:** Replace large, brittle AI skill packages with atomic single-purpose skills that do one thing well. Mega-skills break as context shifts over weeks and months, making them impossible to maintain. Instead, build small composable Lego-like skills and let an orchestration layer string them together dynamically. This approach survives changing workflows and is easier for non-technical teammates to modify using plain English. - **Retrospective AI workflows:** LLMs perform better at pattern recognition over past data than forward planning. Daigle runs daily workflows pulling from Obsidian notes, Slack, Teams transcripts via WorkIQ MCP server, and GitHub to reconstruct what happened across a 3,000-person org. This backward-looking loop — summarize the week, identify what worked, then project forward three to four days — produces more reliable outputs than generative planning prompts. - **GitHub scaling architecture:** GitHub's reliability issues stem from a central permissioning database internally called MySQL One, monorepo growth reversing the industry's multi-repo trend, and CPU constraints from agents multiplying Actions usage. The fix involves migrating to Azure dev compute for fast VM spin-up, breaking out permissioning layers, and rewriting job queuing infrastructure — changes that produce step-change improvements rather than incremental gains. - **Commit volume as AI adoption signal:** GitHub crossed 275 million commits per week in April 2025, putting it on pace for 14 billion commits annually versus 1 billion in all of 2024 — a 14x increase in roughly one year. This growth is linear and still accelerating, driven by agents generating PRs, larger monorepos replacing distributed repos, and a user base that has grown past 200 million accounts. - **Trust signals for open source:** Stars and commit counts are passive, gamifiable metrics that attackers can inflate by aging accounts and submitting cross-repo PRs. A more robust approach involves agentic workflows that evaluate composite signals — accepted PRs across multiple projects, linked social handles older than a threshold, contribution history patterns — letting individual maintainers define their own trust heuristics rather than GitHub enforcing a universal standard. - **Context layer as the missing piece:** Current coding agents lack ambient awareness of business context — meeting transcripts, product specs, email threads, analyst briefings — that human developers naturally carry. Daigle frames this as the next frontier beyond agentic IDEs: a persistent context engine that connects GitHub Copilot to WorkIQ and Foundry IQ so agents inherit organizational knowledge without requiring developers to manually re-explain priorities on every task. → NOTABLE MOMENT Daigle revealed he built an entire revenue planning presentation — pulling from Obsidian notes, Slack, and internal data into a SQLite app — and delivered it to the CFO and CRO without disclosing AI involvement. He deliberately instructed the model to produce visually plain, non-polished output so it would pass as human work. 💼 SPONSORS None detected 🏷️ GitHub Scaling, AI Agents, Developer Productivity, Open Source Security, GitHub Copilot, Microsoft Build 2025