Wow. Thanks for this update. It streamlined a lot of tasks.
Apart from this, next step will be to add suport for building android apps on the android phones itself. No desktop needed.Building on the laptop with agents and installing the build in the phone and testing doea not seem AI native. If everything can run on my android phone, development cycle will speed up.
you already could! just install Termux, npm install your favourite agent harness (pi for one has explicit Termux support, but its AGENTS.md works just fine with Claude Code for example - https://github.com/badlogic/pi-mono/blob/main/packages/codin...), and say you want an android app. It problem solves for a bit, then spits out an apk out to your Downloads folder.
Let me try this. Last year this was a dream. Can't belive we are so close to automate all of this.
My major issue last time was providing the feedback to the agent by running the apk on phone i.e, pass the debug log from the apk back to agent so it can iterate on it without me providing any input.
Also coding agents will happily compile android applications (of maximum complexity) via Github Actions where you can just pick them up with Obtainium. No PC needed
Everything I do for macOS/iOS is already without Xcode but it's a pain in the ass to keep up with changes, and there are things I haven't figured out yet (like AUv3).
This is a good step forward, but keep in mind the claimed gains are about "project and environment setup", not the tasks you deal with on a daily basis in an existing project.
Taking screenshots, optionally with component borders highlighted, and operating the UI with element names like "button1" instead of tap 200,30 looks useful. If I could get it to work.
>Google collects usage data for the Android CLI, such as commands, sub-commands, and flags used. This data does not include custom parameters or identifiable information. This information helps improve the tool and is collected in accordance with Google's Privacy Policy.
You could export BASH_ENV to have Bash processes source a given file at startup.
Zsh has .zshenv, and Fish just has config.fish for everything with the ability to guard certain things within it to login only or non-interactive only.
> How would Google have enough data about a brand new product without collecting that data?
They wouldn't. But on the other hand, they probably have a large amount of in-house Android app developers on whom they can conduct such metrics collection. I wouldn't expect outsiders to have vastly different workflows, because when you get out of the happy path with Android all you get is pain.
Let's see if even mid/big companies with tons of resources, with AI and the right tooling will continue to write webview-apps or, even worse, use some kind of multi target wrapper.
This is great. We also need a tool to expose source jars to agents so they don’t need to compress. There’s a lot of Compose overloads that Claude just guesses at. I built something internally but it needs polish and Claude really struggled with the deep Gradle integration.
You're forgetting the installation ("sideloading", what everyone else calls installation) restrictions they are about to deploy. It will be a significant hassle to install anything without Google's approval. Many F-droid apps are showing warning notices about this upcoming change.
Good, it shouldn't be two clicks for elderly people to install trojans on their phone that then drain their bank account. There should be some explicit confirmation that the user knows what they are doing and they are not being scammed. It is long overdue.
"This APK cannot be scanned and its safety cannot be verified. Learn more/go back" and "learn more" has a link that looks like nothing but is actually a button to actually install the app.
I can think of some easier things, for example popping up a dialog, pressing "install" and having my all actually be installed after that.
You're saying it should look like those damned browser certificate failure sites, with option to open the damn site hidden under button that looks like an unassuming link?
this has been my sort of big tent alignment with AI people. If I'm getting good CLI tooling that _actually works_ (or fixes to existing ones that have been busted forever) then I'm pretty happy.
Things that make systems more understandable to the LLMs ... usually make things more understandable for humans as well. Usually.
The biggest issue I've found is that vibed up tooling tends to be pretty bad at having the right kind of "sense" for what makes good CLI UX. So you still have awkward argument structures or naming. Better than nothing though
It never made sense to me why cars and pedestrians need to share the same spaces. Why can't we have more efficient walking routes that are away from cars?
I've been thinking the same thing lately. It's sorta frustrating that it required bots to force tech companies to make clean simple cli driven development workflows.
It's wild that it took AI to get half the companies on the planet to actually add reasonably priced APIs to their products so I don't have to puppeteer every damn thing with a flakey harness.
> Agents will allow human programmers to get what they've been begging for decades now: proper requirements and flexible, logical, tooling.
...and once this goal is finally reached the programmer will breathe a sigh of relief and then promptly be fired since now the machine can do the job as well as they could.
> Just wait until there are entire classes of vulnerabilities related to LLM usage
This is a valid concern.
There are going to be a new class of vulnerabilities which an LLM is involved which are going to be discovered and it will make it possible to cause catastrophic damage to a company; very easily.
This won't be surprising since we have companies building casual remote code execution tools for "agents" waiting to be hijacked.
I understand that. What about that relates specifically to the Android CLI? That was rafram's question, and mine, and as far as I can tell still hasn't been answered.
I mean, I guess if you're going to say "don't use LLMs", then you also don't want to let agents use the Android CLI, but it seems like raising an awfully general concern in a discussion about a very specific article.
`curl -fsSL https://dl.google.com/android/cli/latest/windows_x86_64/inst... | bash`
The URL shown for individual OSs work, but the script errors for me.
`curl.exe -fsSL https://dl.google.com/android/cli/latest/windows_x86_64/inst... -o "%TEMP%\i.cmd" && "%TEMP%\i.cmd"`
I manually downloaded the exe, but it say socket error. vibe coding is going strong!
Apart from this, next step will be to add suport for building android apps on the android phones itself. No desktop needed.Building on the laptop with agents and installing the build in the phone and testing doea not seem AI native. If everything can run on my android phone, development cycle will speed up.
My major issue last time was providing the feedback to the agent by running the apk on phone i.e, pass the debug log from the apk back to agent so it can iterate on it without me providing any input.
Everything I do for macOS/iOS is already without Xcode but it's a pain in the ass to keep up with changes, and there are things I haven't figured out yet (like AUv3).
>https://policies.google.com/privacy
>Disable Android CLI metrics collection by using the --no-metrics flag.
No thanks, is there no env variable for this? Doesn't Google have enough data already?
How would Google have enough data about a brand new product without collecting that data?
Zsh has .zshenv, and Fish just has config.fish for everything with the ability to guard certain things within it to login only or non-interactive only.
They wouldn't. But on the other hand, they probably have a large amount of in-house Android app developers on whom they can conduct such metrics collection. I wouldn't expect outsiders to have vastly different workflows, because when you get out of the happy path with Android all you get is pain.
F you google. Me too. Why didn't we get a sane way to build android apps before you had to please chatbots?
Is there any step by step process or guidance on it?
There's a link to a repo or you can use the CLI tool to init the skills
I can think of some easier things, for example popping up a dialog, pressing "install" and having my all actually be installed after that.
Things that make systems more understandable to the LLMs ... usually make things more understandable for humans as well. Usually.
The biggest issue I've found is that vibed up tooling tends to be pretty bad at having the right kind of "sense" for what makes good CLI UX. So you still have awkward argument structures or naming. Better than nothing though
The Programmers Brain book was my go to
http://www.robertames.com/blog.cgi/entries/the-unix-way-comm...
TLDR: standard list of long options <https://www.gnu.org/prep/standards/standards.html#Option-Tab...> and short options from -a to -z <http://www.catb.org/~esr/writings/taoup/html/ch10s05.html>.
...and once this goal is finally reached the programmer will breathe a sigh of relief and then promptly be fired since now the machine can do the job as well as they could.
> Just wait until there are entire classes of vulnerabilities related to LLM usage
This is a valid concern.
There are going to be a new class of vulnerabilities which an LLM is involved which are going to be discovered and it will make it possible to cause catastrophic damage to a company; very easily.
This won't be surprising since we have companies building casual remote code execution tools for "agents" waiting to be hijacked.
I mean, I guess if you're going to say "don't use LLMs", then you also don't want to let agents use the Android CLI, but it seems like raising an awfully general concern in a discussion about a very specific article.