The developer workflow is being rebuilt so agents can keep up
GitHub's session protocol, Google AI Studio's instant Android builder and a filesystem that cuts checkouts from 36 minutes to seconds. The dev loop is being rebuilt for agents.
- AHP is an open standard for agent session continuity, designed to tolerate network interruptions and reconnections, with libraries in Rust, TypeScript, Kotlin, Swift, .NET, and Go. Self-reported m:…2424
- Google AI Studio builds full-stack Android apps from a single prompt, with Cloud Firestore, Firebase Authentication and Firebase AI Logic built in. Self-reported m:…6835
- ABFS cut Android checkout times from about 36 minutes to under 8 seconds and full build times from about 80 minutes to just over 2 minutes. Self-reported m:…0183
- The Android 17 open source release contains 1,084 Git repositories and 1.62 million files, and a built workspace needs 300 GB of storage. Self-reported m:…0183
Sessions that survive
GitHub's AHP is an open standard for agent session continuity, built for a simple frustration: an agent session should not die when the device changes or the network drops. It is designed to tolerate network interruptions and reconnections, and it uses a reducer-based state object to drive client logic. Every AHP server must implement root, session, and chat channels, and AHP uses URI-identified channels as the coarsest unit of functionality. m:…2424
The implementations show the intended reach. VS Code's AHP implementation is written in TypeScript, GitHub's is written in Rust, and AHP libraries are available in Rust, TypeScript, Kotlin, Swift, .NET, and Go. The background is worth noting: GitHub Copilot was originally built as a VS Code extension with a bundled harness, and VS Code extension lifetimes are tied to the window lifecycle. AHP is the fix for that history, the session outlives the window. m:…2424
Apps from a single prompt
Google AI Studio now lets developers build full-stack Android apps from a single prompt, with the backend included. It integrates Cloud Firestore for backend services, Firebase Authentication, and Firebase AI Logic, and it includes an HD emulator with audio support and visual screenshot debugging tools. When the app is ready, it allows direct publishing to a personal Google Play library. The pitch is that the distance from idea to a working app with a real backend is now one prompt. m:…6835
Repos at agent speed
None of that works if the repository itself is slow, and that is the problem ABFS attacks. ABFS reduced Android checkout times from approximately 36 minutes to under 8 seconds, and it reduced Android full build times from approximately 80 minutes to just over 2 minutes. Shared cache reuses up to 100% of build actions across CI and developer workspaces, and ABFS enables parallel agentic workflows across multiple workspaces and agent orchestrators. m:…0183
The scale explains why it matters. The Android 17 open source release contains 1,084 Git repositories and 1.62 million files, and a built workspace requires 300 GB of storage. Hardware vendor code and first-party applications increase the size of a typical Android source tree by 2-3x. Fetching AOSP source code traditionally took over 30 minutes, and building it traditionally took over an hour on highly capable developer workstations. ABFS uses content-addressability across the entire stack to lazy-load and cache source code, developer workspaces, intermediate build files, and build results. It is a full suite of software comprising a Linux kernel filesystem module, user-space daemons, and server-side infrastructure, plumbing built so agents can work in parallel without waiting on the repo. m:…0183
What it adds up to
Put the three releases together and the developer's loop changes shape. Sessions that survive a network drop, apps that build from a single prompt, and repositories that check out in seconds are not three separate conveniences. They are the three preconditions for letting agents do development work without a human babysitting each step: the agent needs somewhere to keep working, something to build with, and a codebase that answers fast. The vendors are not selling smarter models here. They are selling a workspace that does not get in the model's way. m:…2424 m:…6835 m:…0183
The practical test for a team is simple: time how long an agent waits. If the wait is the session dying, the fix is a protocol like AHP. If the wait is the repository, the fix is a filesystem like ABFS. If the wait is scaffolding a new app, the fix is a builder like AI Studio. The three releases agree on one thing: the agent is no longer the slow part of development. The workflow around it is, and that is what is being rebuilt. m:…2424 m:…0183
What is still unresolved
- The AHP post does not say how many hosts or clients have adopted the standard.
- The ABFS figures come from the team that built it; no independent measurement is cited.
- The AI Studio post does not say what the generated apps cost to run.
Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.
Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.