Insights on XR, Android CLI, Compose and everything in between
Overview
Attending the Google Android Dev Summit in Mountain View was an exciting experience given it was my first time attending an official event at Google HQ. Compared to massive, spectacle-rich conferences like Google I/O, and Cloud Next, the scale here was noticeably more intimate, allowing direct technical conversations with Google’s core platform team and offering unique insight into shifting engineering priorities at Google — more on that to come.
The primary focus centered on enabling agentic coding paradigms, by introducing BYO model capabilities in Android Studio, and expanding command-line tooling via the Android CLI, while also enabling richer and more performant user interfaces with Compose/A2UI improvements.
Being there in person was insightful as it provided additional context on upcoming architectural and cultural shifts in the Android ecosystem, revealed by what was said, but more importantly, by what was left unsaid. This was particularly evident with Android Studio: while it received a minor update, it clearly took a back seat to the Android CLI updates and seems to be falling out of favor with Google as the preferred developer experience.
Android CLI & Agentic Workflows
The introduction of extended command-line interface (CLI) support for Android represents a major cultural shift towards agentic automation. Built specifically to improve execution performance and minimise context token usage, Android CLI allows autonomous AI agents to execute code development, build tasks, run tests, and confirm they have satisfied the requirements in a verification loop, with the simplicity and performance of the terminal.
A key insight emphasised during the sessions was the importance of hand-crafting your repository’s agents.md files rather than delegating them entirely to AI generation, ensuring strict project standards and precise instruction guardrails. By hand-writing the documentation, AI agents are more likely to follow it as expected. This was particularly interesting to me, as it highlights the inherent risk of relying on AI to define the rules while also being the one to execute the task.
Full video from Ben & Steven: Agentic Engineering with Android CLI
Agentic development often struggles when agents require human intervention for constant, trivial choices. Too often, I’ve set an agent to a task, stepped away, and returned 15 minutes later only to find zero progress because it stalled on a basic naming decision. A proven alternative in the agentic SDLC is the ‘grill me’ approach: where you prompt the AI to ask all clarifying questions upfront, then execute the task autonomously. Paired with self-verification loops, this eliminates constant hand-holding and enables material productivity gains.
Another insight was that agents are often far too confident in their own abilities and finish a task thinking it is complete when it’s really not. Just think how many times have you received an answer from an AI, asked it to double-check its work, only to watch it immediately backpedal.
From Google’s experience, asking the agent to verify and show you the requirements are met, materially increases the reliability of the outcome. This inherently increases the runtime demanded given the longer verification loop, but generally provides better overall results as there’s less re-work required.
To summarise, by resolving any ambiguities early on in the SDLC workflow, it allows us, as experienced engineers, to focus on other, more nuanced items while the agent is running in the background doing what it does best. This is where we can add real value as humans and provide a differentiating factor to agents.
To explore these developments further, check out the official presentation and developer documentation:
Full Presentation: Agentic Engineering & Android CLI
Developer Guide: Android CLI Tooling Documentation
Standardising Best Practices with Android Skills
AI agents often fail to adhere to Modern Android Development (MAD) patterns, frequently generating legacy boilerplate, outdated patterns, or outright hallucinating. This does align with my experience working on both Android and .NET APIs, where agents are generally much better at generating .NET API code over Android code. Though this is not surprising given the Android ecosystem evolves so quickly, models often lack the volume of training data found in more stable environments like .NET.
To solve this, at ADS, Google announced over 30 official Android Skills to set engineering standards for code-generation agents. These skills specifically target:
- New functionality: Defining a knowledge base for recently released features that AI hasn’t been trained on yet.
- Complex implementations: Guiding hardware integrations like advanced camera APIs that are specific to Android.
- High-risk areas: Android Security protocols & Play Store compliance.
These new announcements demonstrate that Google acknowledges the current ecosystem has not been favorable to agentic engineering, and is now providing flexible, AI-friendly standards so developers can adopt what works best for them.
Next-Generation Android Prototype Tooling
Another standout revelation was an upcoming, not-yet-named rework of the Android Emulator that runs completely differently.
Engineered to address performance limitations, this next-generation Android ‘simulator’ (currently just called ‘prototype’) provides faster boot times, significantly lower system resource consumption, and smoother runtime execution compared to traditional Android emulators.
This leap in performance is critical for local agent development loops and high-throughput automated testing runs. And allows agents to verify their work much more efficiently.
Anyone who has been working on Android for some time knows how big a leap this is and it was one of the most talked about changes at the event. Everyone was excited to be able to run and test their apps quickly, especially if you work on enterprise sized apps — this is a game changer.
Hands-On with Android XR & Spatial Computing
Hands-on workshops provided direct real-world time with emerging Android XR devices and spatial app development workflows like the newly announced XREAL Aura.
These workshops demonstrated how the ecosystem separates light, head-mounted glasses from an external computing ‘puck’ that handles primary processing, enabling extended screen mirroring and multi-window productivity layouts.
The current trajectory seems to focus on productivity, resting media consumption, and hands-free utility rather than continuous, all-day wear.
While I did feel a little nauseous wearing them for longer periods (though given the limited headsets available, I had the wrong size for me so that likely contributed), they were impressive in their presentation ability and hand tracking. This made the experience feel quite intuitive and became quite ‘routine’ after only a short amount of time — a good sign new tech will become normalised to people.
It still remains to be seen whether they will get large-scale user adoption, but I can see this generation of AR fulfilling a niche in the market more than the full VR headsets, or the glasses-only options which have significant privacy concerns.
Conclusion & Outlook
The Android Dev Summit made it clear that Android development is rapidly evolving to prioritise AI-driven, agentic development workflows. By extending support for the Android CLI, standardising Android Skills, and experimenting with more efficient virtualisation tools, Google is establishing a decisive lead in modern agent developer enablement.
A2UI is also being enhanced to provide a pattern for dynamic, agentic experiences, right in the palm of your hand. This is where companies will be able to differentiate themselves from their compe*****s with richer and smoother experiences. More details: Agentic UI with A2UI
For mobile engineers and technical leaders, adopting these tools is no longer optional, but critical for supporting high-performing engineering teams, shipping standout user experiences, and remaining competitive in a rapidly evolving ecosystem.
Android Dev Summit 2026 was originally published in ProAndroidDev on Medium, where people are continuing the conversation by highlighting and responding to this story.
