DPE Summit 2027 is approaching fast!

Our 5th DPE Summit will be hosted in San Francisco, California on February 2-3, 2027, and we can promise one thing: you don't want to miss it! Two packed days of talks from the biggest companies in the world on all things DPE and DX in the AI era.

Think you have an interesting talk to share? The DPE Summit Call for Papers is open now through the end of the month!

Early Bird tickets are on sale now for $349, don't miss your shot to attend!

Isolated Projects graduates, and IDE sync gets a lot faster

Gradle 9.7.0 shipped on August 6, promoting Isolated Projects from experimental to incubating. The feature enforces boundaries between projects in a multi-project build, allowing Gradle to configure them in parallel rather than one at a time. On Gradle's own 300-subproject build, IP reduced the median IDE sync time from 84 seconds to 47 seconds.

The catch is that Configuration Cache is a prerequisite, and the isolation constraints are real: build logic in one project can no longer reach in and mutate another project's state or the build's mutable state directly, so calls like project(":other").tasks now fail fast instead of quietly working. Task graph discovery across projects remains sequential in 9.7.0, so the CLI speedup is smaller than what the IDE sync sees.

The Gradle team's own migration is worth copying. They leaned on convention plugins, so each project applies only the build logic it needs, and nothing reaches into another project's state. They rolled the feature flag out gradually, keeping it explicitly disabled on most CI jobs at first via a command-line flag that overrides whatever is checked into gradle.properties, while deliberately keeping it enabled on at least one small CI job so the constraints stayed validated throughout.

Why Ramp built its own coding agent instead of buying one

Gergely Orosz's Pragmatic Engineer sat down with Ramp's CTO Rahul Sengottuvelu, head of engineering Hamid Dadkhah, and principal engineer Zach Bruggeman to talk through Inspect, the fintech's homegrown AI coding agent. The short version of why they built it instead of adopting Codex, Claude Code, or Cursor: engineers wanted to run far more agents in parallel than a local machine supports, needed stronger frontend tooling, and kept running into demand for remote development environments that off the shelf harnesses didn't handle well.

Inspect runs on remote sandboxes with access to most of Ramp's internal data sources and verifies backend and frontend changes on the machine itself. It's built on React and Vite, with Cloudflare Durable Objects and SQLite for state, the Cloudflare Agents SDK, and Modal for sandboxing. According to a separate InfoQ report, it now powers roughly 30 percent of merged pull requests across Ramp's frontend and backend repositories.

Ramp isn't alone. Block has its own agent called Goose, which is open source. Stripe calls its version Minions, and Shopify's is River. A small but growing list of companies are deciding the harness itself is worth owning.

Gradle launches a security subscription for the build tool itself

Gradle's open source end-of-life policy only guarantees fixes for the latest minor release of the current and previous major versions. Teams stuck on an older major or behind within the current one, either upgrade to get a patch or wait for the next scheduled release. The new Gradle Security Subscription adds three guarantees on top of that:

For organizations that cannot migrate to the latest version of Gradle or have a formal compliance obligation, the Gradle Security Subscription offers a new option.

Nothing changes for teams that don't subscribe; the open source policy stays exactly as it was.

The reasoning is that the build tool is part of the attack surface, not just a place where dependencies are pulled from; a point the post makes by way of XcodeGhost, SolarWinds, Codecov, s1ngularity, and Gradle's own 2023 wrapper attack.

Gradle Is Going Agentic

In an August 19 post, Gradle's Tom Tresansky and Laura Kassovic argue that coding agents are already part of the build loop, whether Gradle invited them or not. More and more people running Gradle every day aren't typing ./gradlew build themselves; they're asking an agent to run the build, paste in the failure, and fix whatever broke.

The response is a project the team calls Agentic Gradle, split into two parts. The first is official Gradle Skills, small, focused instructions that teach an agent how to do something specific with Gradle. The second part is a benchmarking tool that will determine whether the skills, or any other Gradle features created by this initiative, are actually worth it. These features are aimed specifically at the places where agents make today's build tooling awkward, such as the build output, the CLI, and the documentation.

Getting there came with a real tradeoff. The team is holding off on making Configuration Cache the default in Gradle 10 to prioritize this work instead. There's a new #agentic-gradle channel on the Gradle Community Slack for anyone who wants to follow along.