Honestly why not run OpenClaw? There are ways to run it safely and it's kinda built for that use case
I haven't blogged about that work specifically, but the main success differentials were providing the agent with everything needed to inspect potential issues (view hierarchy, overlapping views, occluded views, rotation, insets, camera cutout) and the entire navigation graph.
I've used both with the 4.5 anthropic models and Junie gets in the way. I'm not trying it again
Make it viral.
I don't see this on pocketcast yet - are you still publishing there or is it just delayed?
And raisins, tea, and coffee too. Every dog is different, sister in law's ate an chocolate cupcakes and was okay 🤷♂️
Grapes are another big one. Simply can't allow our kids to have them in the house because they just don't understand.
Wonder what Atlassian bases their status pages on.
Funny what breaks down when internet services fall apart. Seems like their payment system is based on Toast status.toasttab.com but their status page is at least accurate
My kids like to decorate and we just let them pick a room and put up banners, streamers, and decorate - could be fun to try? Otherwise think you'll have to vibe code one. Totally get what you mean, gaming seems rampant with that stuff
Time to vibe code a browser?
I've found that I'm able to use it as a workflow assistant to constantly apply deep analysis - mostly in what it can observe via UI, but there's potential with database, navigation, etc. Will be updating docs and releasing the next version when I can. Feedback welcome!
I'd hang, just need to put kiddos down
And if you want to try it out I just released v0.0.5
zillow.github.io/auto-mobile
Awesome discussions! Was refreshing and neat to see panels done this way
It's going to be waiting a long time with SO traffic nosediving
Not in the same way I'm talking about. They have no context for build system, language server, your project dependencies, and what would actually compile. Tools that provide that context would make those agents better.
I've been thinking about this since I'm creating tools to do it for UI testing. Rather than just letting the AI monkey around with the function inputs, what if we had an MCP that could explore source code and interfaces the way I've been able to explore UI?
AutoMobile is OSS on GitHub! Come check it out at zillow.github.io/auto-mobile #LiveDemo #AndroidDev #dcnyc25
I'll be at Stage 2 at 3:30pm, come with questions and take some stickers
AutoMobile talk today at #dcnyc25 #LiveDemo come check out the tooling I built to make #AndroidDev automation automated
Metaprogramming with @anthonycr.bsky.social 🎉🙌🤠
Programming programs that program programming programs #AndroidDev
@stacydevino.bsky.social bit.ly/notandroiddev definitely the keynote we needed for 2025
Droidcon videos typically get posted a month later
Awesome talk about state of #AndroidDev, Shift-Left. Agree with everything about needing to lean into skillsets to make it in the coming years
Android MCP space is about to 🚀
Sure thing 🙌 likewise happy to share yours. I also have an MCP talk on day 2 about a new project called AutoMobile.
I created an Android MCP SDK, hoping it can be transferred to the MCP GitHub org and continue to iterate. I see a lot of potential capabilities with it, even in production eventually. www.github.com/kaeawc/andro...
