Develocity (Gradle)
Day 2 after signup
UBC, University of Zurich, Netflix, Develocity at DPE Summit
DPE Summit 2027 announces its first speakers. Plus a new whitepaper on the AI-native SDLC, why slow builds and flaky tests cost you twice, two companies rebuilding Git's backend, and new GenAI collaboration research.
Why it works
- A numbered overview at the top gives readers a quick sense of the issue’s range before they reach the full stories.
- The email puts the timely event and ticket offer first, then follows with substantive summaries so readers can choose which topics to explore.
- Each article section explains the central idea before presenting its call to action, making the links useful and specific.
How it is built
- Numbered topic overview
- Summit announcement
- Ticket offer and CTA
Email copy
01 DPE Summit 2027 announces its first speakers 02 A new whitepaper models the AI-native SDLC 03 With agents, slow builds and flaky tests cost you twice 04 Two companies rebuilt Git's backend in the same week 05 What 131 developers actually say about working with GenAI 06 DPE & AI job openings DPE Summit 2027 announces its first speakers Mark your calendar for DPE Summit 2027, February 2-3 at the Midway in San Francisco. We're bringing together over 600 developers, platform, build, and CI engineers, and the engineers who run coding agents at scale for two full days of practical insights, deep dives, and lessons learned from some of the most influential companies in the industry. What's on the agenda 50+ speakers across 30+ sessions, from the teams furthest along on the delivery side of the AI-native SDLC: • DPE Summit Opening Keynote: Hans Dockter, founder of Gradle Build Tool and CEO of Develocity, kicks things off on day one. Session details coming soon. • One Team, Multiple AI Workstyles: Why GenAI Adoption Isn't One-Size-Fits-All: Marie Salomon, Thomas Fritz, and Gail Murphy of UBC and the University of Zurich share four developer personas that split work between GenAI and teammates, and how leaders and DevEx teams can design for those mixed workstyles. • Tag. Publish. Hope.: Aubrey Chipman and Roberto Perez Alcolea of Netflix share how to keep AI-generated library changes from breaking the services that use them, plus a five-part checklist for AI-ready library ecosystems. Early Bird Pricing: Today's the last day to get your ticket for $349. Rates increase to $449 after Friday, November 13. Get your ticket The AI-native SDLC: from delivery-limited to idea-limited Hans Dockter's new 45-minute article starts from a blunt diagnosis: coding agents have collapsed the time it takes to write code, but delivery hasn't kept pace. Most SDLCs today are delivery-limited, the bottleneck is verifying and reviewing what agents build, not writing itself. The paper models the AI-native SDLC across eight stages, from Intent to Maintain, and goes deepest on Coding, Code review, and Release, the three stages where verification decides whether delivery can keep pace with agents. It also introduces two loops that sit alongside the stages: a verification health loop that watches for flaky tests and slowing builds, and an eval loop that tests the coding agents themselves whenever their model or harness changes. The destination Hans argues for is an SDLC that's idea-limited instead of delivery-limited, where an organization ships as fast as it can decide what's worth building, not as fast as it can verify what agents already wrote. Read the article With agents, slow builds and flaky tests cost you twice Slow builds and flaky tests have always cost developer time, and agents haven't made that cost disappear, they moved it. Now you pay it twice: in compute and tokens an agent burns waiting on slow builds and chasing flaky test failures, and in the time developers spend waiting on the agent. Then there's drift. Main keeps moving under every open change, and that alone caps how many agents you can run in parallel. When more failures land than the few people left can handle, the fleet slows to human pace. Most changes have to merge without a human touching them, and only the toolchain can get you there. The fix is to treat your toolchain like a production system. The full post covers the three changes that cut costs, and a simple way to put a price on your false build and test failures. Read the full post Two companies rebuilt Git's backend in the same week On November 2, OpenAI published "Scaling Git at OpenAI," detailing Pando, a new backend behind its internal monorepo. Three days later, GitHub published its own engineering post on Project Rubicon, a new Git storage architecture it tested handling 5,000 concurrent coding agents on a single repository at 7,200 pushes per second. Neither team set out to replace Git. Both