Keynote: Is the Developer Lifecycle Dead?
summary
Argues that agentic development produces far more code and faster — agents run parallel threads overnight, founders set 2 AM alarms when loops stall, and teams see a Sunday idea become a shipped iPhone app. Recommends storing session logs with code so the “why” travels with commits, treating delivery as loop engineering with long-running agent runs and evals, budgeting tokens (example cap $500/month) instead of dev headcount, and building a decentralized Git network with local mirrors to serve always-on agents at global scale.
AI-written summary · about
chapters (30)
- 0:00Is the Software Lifecycle as We Know It Dead?
- 2:36Guest Intro - Thomas Dohmke, ex-GitHub CEO, now building Entire
- 4:09What Agentic Development Breaks in Systems Never Built for It
- 5:11From Single-Threaded VS Code to Running an Orchestra of Agents
- 6:12Every IC Is Now Also an Engineering Manager - of Agents
- 6:43Why You Want Agent Loops to Run Longer, Not Return Every Three Seconds
- 7:14Shipping Code You Barely Read - and Not Giving Up Control
- 8:15Code Review Was Always Flawed - The Spot-the-Difference Analogy
- 10:15Agents Writing Code Overnight That References Code You Never Saw
- 10:47Moving From Human Code Review to AI Reviewing AI
- 11:47Managing the Agent Arena: Codex Fixing What Claude Found
- 12:17Is There Still a Role for the Developer at the Merge?
- 13:19Ideas Still Come From Humans - Machines Won't Define How We Live
- 14:52Session Logs Are as Important as Code - Capturing the Why
- 15:24Agents That Tell You Honestly They Did the Wrong Thing
- 17:26How the Inner and Outer Loop Change in the Agentic World
- 19:32Wanting a Slow Turnaround: Letting Agents Work While You Sleep
- 20:33Founders Setting 2 AM Alarms to Check on Their Agents
- 21:34Designing Loops So an 8-Hour Run Doesn't Produce Garbage
- 22:36Loop Engineering in the Wild: Who's Actually Doing It
- 23:38Why the FOMO Is So Strong: Sunday Idea to Shipped iPhone App
- 25:13Selling Copilot at GitHub Was an Uphill Battle - Now It's Normal
- 27:16Token Maxing: How Should an Engineering Leader Think About Tokens?
- 28:47What Happens When an Engineer Hits Their Token Budget?
- 30:19Are Tokens Tied to Revenue? Features vs Business Outcomes
- 32:23Where Not to Burn Tokens: Use a Test Framework, Not an Agent Reading Logs
- 34:25Why Tokens Aren't Like Cloud: COGS vs OPEX for Engineering Leaders
- 35:27What Does the Next Platform - the Next GitHub - Look Like?
- 36:29Thesis Two: The DevOps Lifecycle Becomes Loop Engineering
- 37:31Thesis Three: A Decentralized Git Network for Agents That Never Sleep
