top of page


20 NOVEMBER 2026
BANGALORE


( SPEAKER )
Mansi Tanna
Senior Mobile Developer
Wurth IT India
( SESSION )
The world's premier event for Android developers.
Most AI features stop at giving advice or summaries. The real friction begins when your agent needs to take action and write data to other apps like Calendar, Messages, or banking tools.
Android developers are currently wrestling with how to handle these automated actions safely. Should you hand off writes using standard system intents, or execute them directly? How do you manage permission budgets without killing the user experience?
In this circle discussion, we skip the basic tutorials and share what actually works in production. We will trade real-world safety patterns, discuss what causes user friction, and debate where teams draw the line between on-device and cloud processing. Bring your agent integration war stories, permission headaches, and architectural nightmares. This is a practical pattern swap, not a lecture.
( SESSION )
The world's premier event for Android developers.
Smooth playback is one of the biggest factors behind a great feed experience, yet many Android apps still show a spinner on every scroll because media preparation only begins when the user arrives.
The obvious fix is preloading. DefaultPreloadManager prepares media before it is needed, and we'll cover how it works, how to integrate it into an existing Media3 player, and what it costs in bandwidth and memory.
But preloading alone breaks a feed. Every ExoPlayer instance holds a hardware decoder, and a device only has a handful. Create one per feed item and playback stops working somewhere around the sixth scroll, usually not as a clean crash but as black frames and a codec exception several items later. That's where PlayerPool and rememberPooledPlayer, new in Media3 1.11, come in: a fixed set of players recycled across a feed of any length.
We'll build it live. A naive feed scrolled fast until it breaks, so you see the failure mode as it actually presents. Then preloading, with time-to-first-frame measured before and after on a mid-range device. Then pooling, with the memory profile alongside.
The last section is tuning, because the defaults are not answers. Pool size and preload window trade against memory, decoder pressure and the user's mobile data. We'll look at what changes between a 3GB device and a flagship, what happens when someone flings through forty items in two seconds, and how to pick numbers you can defend in review.
bottom of page