Over the recent break, across four or five sessions, I wrote a set of prompts around ~500 words in total.
The result was Claude scanning my network for active ports using nmap, fuzzing those ports with cURL, documenting its findings, self-directing web searches for API/SDK docs for my Hue bridge and ancient Samsung TV, then building a set of scripts to control my lighting system and a fully functional HTML+JS remote for my TV.
The most entertaining part was Claude prompting _me_ to pop into the living room and press the button on the Hue bridge so it could fetch an API key.
The most valuable part? The understanding I gained secondary to generative act. I now understand the button on a Hue bridge literally just tells the device to issue a new API key at the next request. I understand how Entertainment mode works, and why. I understand how Samsung SmartThings is mediated via websockets - and just how insecure decade old Samsung TVs are.
Around 500 words to gain all this? I hate to buy into the hype, but it feels inflectional.
Worth pointing out I'm not SSHing to a different device. Claude Code installed and running directly in Android terminal on my phone.
I've built ASP.NET Core APIs on-device this way. Install dotnet in the terminal and Claude can write code, build, run unit tests, and even run the API on localhost. Then use `git` and `gh` to commit, push and raise a PR.
As long as you figure out how to verify that the built thing actually does what it's supposed to, ideally with automated tests, it's almost fire-and-forget if you're good at explaining what you want and need.
As for the local env, I'm treating the Android terminal as a sandbox. Anything gets trashed I just reset and reinstall my toolchain.
I won't pretend I'd use this workflow for anything high-stakes. But for simple things like "I wonder how my Hue lights actually work?", its viable.
So, I do this when I am sitting on the couch and too lazy to boot up my laptop that I normally do work on, but it never gets much further than updating, pulling or pushing one or two containers, or more times than not trying to remember what port AI have something on so I can connect the companion app to it.
It's not a bad idea in full, but "death coding" is a ridiculous notion.
That said, this would be interesting to someone who didn't know these tools could be stitched together in this way. I think that's a big part of why it's on the home page.
I'm new to hacking (come from an electrical/nuclear engineering background but never did much with software). For reference, just learned what postgres was 2 months ago.
Took a lot of tinkering to figure out but that's more a skill rather than complexity issue. Working from a laptop is certainly better, but was able to get good amount done (like building v1 of a backend and setting up a cloudflare tunnel for a PC) on a long bus ride where I would've gotten side eyes for using a laptop.
I'm no doctor but I'll bet "Doom Coding" is still not healthy but it's better than doom scrolling on X.
Thank you for the comments! I've been learning from these threads (Like tmux or dropbox article lore)
I haven't set up a vibe coding phone environment, nothing has stopped me at all as I agree it is really simple/"basic", but this post made me actually go do it.
I know this is probably in jest, but when someone invites you to a party it's not because they just want your atoms in the same room as them.
In regards to doom coding: I would chop off my arms before coding/prompting on a phone. Also, think about your cervical, neck etc! I feel like I'm taking crazy pills!
I am just hoping they actually took a break from doom scrolling while driving as then at least I can say I had some non zero positive impact on their lives.
5 years phone-free and I do not miss it. People use them as security blankets to avoid having to be present for more than 5 minutes at a time with other people or even just exist in their own heads. I now find this behavior immature and gross but avoiding it would mean not having friends.
A smartphone is like toilet paper. No one wants to watch you use it.
There is a bar like that where I go sometimes, it is in a cave, some people got Wi-Fi from the staff, and you have some reception if you stand near the front door, but it is mostly a network-free zone and it is great.
Another thing we did from time to time at the restaurant is to put all our phones stacked in the middle of the table, anyone who picks up his phone before the end of the meal for any reason pays the bill for everyone. So far, no one did.
When I leave home for less than a day I pack no electronics of any kind and enjoy the peace in my own head to think about the next problems I want to solve in my universe.
I pay with cash exclusively in public so tap and pay is not an issue. If I ever need to be reachable for emergencies I can carry a pager but so far this has not been worth it.
I put that in the same bin as all the “Stop doomscrolling” apps. You can’t prevent doomscrolling by adding another app on your phone. Get rid of the phone (and all other screens), one does not need to be able to look up everything in a moments notice. Write it down on a paper and do it later.
At worst, put your ideas into a notes app and then go back to where you are... this is just anti-social and borderline psychotic imo.
The UI isn't as good as a laptop but maybe it's all my years of swiping, liking, and navigating between apps. In a very sad and concerning way, phone time feels like home.
- Email Claude to start the coding
- Claude emails you with any thing it needs acked on.
- you reply back to emails telling it what to do.
- maybe Claude can run your program and send back screen shots.
seems easier then getting a vpn working. What is the downside to using email?
No.
> What is the downside to using email?
Email is clunky and feedback is not immediate.
> seems easier then getting a vpn working.
Tailscale is easy for a dev to get going and very reliable. The author uses the Termius SSH app with Mosh, so it keeps the same SSH session going across device sleeps and disconnects. Tmux is helpful, too.
I do exactly what the author is doing, except I use a $5 Linode VPS, instead of a Mac at home.
He doesn't seem to be credited on this page, but I believe Pieter Levels (@levelsio) actually popularized this scheme. The author documents a nearly identical scheme.
It's not such a crazy idea.
You're vibe coding on a smartphone into an external computer. You already abandoned "Immediate feedback" and "cohesion".
It usually comes from the bad experience or poor exposure.
Its hard to hate on them when it comes from a position of limited exposure.
The idea is to create a modern "terminal"[0]
My main computer is a Macbook Air, which I carry around with me. It's purpose is for: internet, using Microsoft products when I'm forced to, Zoom/meetings, and SSH (or Mosh).
Most of my work is not done on this Macbook, instead I use it mostly as a terminal. I have a desktop that's sitting behind my TV so that it can be my TV and gaming system (yeah I know Monitor > TV. I'm a filthy casual and I don't care). I have a mouse connected to that computer and instead of using a keyboard I use ydotool (Wayland xdotool) with a shortcut on my iPhone or a script on my android phone or from my Macbook. I don't have to get up from the couch and I don't need a clunky wireless keyboard to clutter my livingroom.
Additionally I have a few pis and an old android phone with Tailscale installed on them. That's come in handy before as a machine's been disconnected and so I couldn't ssh from outside. Also makes it really easy to do a jump if you want to keep a machine off Tailscale or you don't have full control (like my 3D printer).
This setup is very natural feeling if you live in the terminal. I actually started doing this when I started doing HPC work. In a setting like that you're never sitting in front of the computer you're doing most of your work on so it kinda clicked "why was I restricting myself outside work?" Plus there's the side benefits of I always have access to my media, tools, and other stuff. You can do exactly the same thing with a phone but I like having a keyboard and the air is very lightweight and has a long battery life. Any netbook would have done the job tbh.
[0] There's a reason they're called "terminal emulators" rather than just "terminals".
I've done this for decades. screen or tmux (although I still confuse the keybindings between the two).
When coding on the move (mostly when I had a long commute or was away from the office visit clients) I'd use the Linux console (Ctrl+Alt+F1-F6) rather than X.
Even in the office I had an old amber/green terminal that connected to my Linux desktop via a serial port.
Nowadays I have a 14" USB-C monitor (ASUS Zenscreen) that sits beneath my main monitors which runs a terminal full screen.
It definitely looks cool and I might give it a try, but I do love my dark mode and color syntax. My understanding is that color on eink is pretty limited. It also isn't worth it to me if I'm going to be spending $500+ on a "monitor". But I'd definitely love if things moved in that direction.
Honestly if Apple wasn't so insistent on making the iPad not a general purpose computer I'd use that as my thin client.
I have no interest in LLM, or vibe code. Even though I miss the capabilities of intellij, nvim can fill the roll in the terminal very nicely, except rust analyzer filling up storage fast,
I also have a spare mobile, which I use to wake the computer up. And I have a python script running on it, to shutdown the computer in case of power failure.
After initial hiccups it working pretty well, except cats turning off the router, well how many can use the excuse that I couldn't finish the work because cat controls your network. LoL
Also, I heard that you can install Tailscale on it[0], so that can act as a gateway which is nice.
Yes. tmux is essential. It's great to be able to monitor a session from desktop, or to pick up an existing conversation i'm having on the computer on my phone. In my shell, I have gemini flash wrapper that I use to manage my sessions so I can quickly connect to an existing one, or create a new one with natural language.
> He doesn't seem to be credited on this page, but I believe Pieter Levels (@levelsio) actually popularized this scheme. The author documents a nearly identical scheme.
I've been doing this (tailscale + termius + tmux + ssh) for at least a year and a half. First with Aider in this exact setup, and now with Claude Code and Codex.
Sorry if anyone, especially the author, took it this way.
so it really sort of falls back to what you're doing with the llm. code maintenance isn't novel development, which isn't polishing.
and all of them mentions Tailscale. I would not be surprised if we hear in a few days it got next big fund and all of this is just a preparation for it
OpenCode has a webUI, you can simply host that on your machine at home and VPN to it.
https://opencode.ai/docs/server/ (sadly no screenshots, but its a pretty good GUI, looks like their desktop app)
Far more resilient and performant than a web client.
If it disconnects you just reload the page. It can work just fine in the background because it’s not running on your phone.
Just like you can refresh the ChatGPT website, but OpenCode lives on your pc at home, not OpenAI servers.
I really like Termius, have you tried it? I think I tested out Blink when I was trying various SSH/shell apps and chose Termius over it, but it’s been so long now that I completely forget why.
EDIT: does Blink give you a local shell as well like vs only SSH/mosh?
> The opencode serve command runs a headless HTTP server that exposes an OpenAPI endpoint…
Unless I missed it, there’s no mention of a web UI?
Thats exactly the approach I took with https://github.com/cloud-atlas-ai/miranda, Telegram bot, PR is the human review point, tests + CodeRabbit catch most issues.
Bot intercepts Claude's AskUserQuestion calls via a hook, sends me an inline keyboard, injects my answer back into the session. Claude keeps working, PR still happens—but I can unblock it from my phone in 5 seconds instead of rejecting a PR based on a wrong guess.
Claude is making it easier to have bespoke data and dashboards for anything. We're going to make a lot of them, for all reasons. I've also made apps with Django interfaces, but quick, global interfaces are going to become in demand.
The gap I wanted to fill: when Claude is genuinely uncertain ("JWT or sessions?" "Breaking change or not?"), it either guesses wrong or punts to the PR description where you can't easily respond.
Built a Telegram bot that intercepts Claude's AskUserQuestion calls via a hook, sends me an inline keyboard, injects my answer back into the session. Claude keeps working, PR still happens—but I can unblock it from my phone in 5 seconds instead of rejecting a PR based on a wrong guess.
Works in tandem with a bunch of other LLM enhancers I've built, they're linked in the README or that repo
You can connect Linear to Cursor's web agent, which makes Linear issues assignable to the agent directly and kicks off Cursor's take on remote coding agent. You can then guide it further via Cursor's web chat.
If Claude Code on iOS supported Linear MCP (as it does on desktop), you can run a similar issue handoff to agent to issue update workflow, albeit without direct issue assignment to the agent "user". Easy to use labels aka tags for agent assignment tracking, as well.
For my hobby projects, I've been using Linear + agentFlavorOfTheMonth quite happily this way. I imagine Github issues, Asana, whatever could be wired up in place of Linear.
I'm just starting out with building with Claude - after a friend made this post he sent me a Steve Yegge interview (https://m.youtube.com/watch?v=zuJyJP517Uw). Absolutely loved it. I come from an electrical/nuclear engineering background - Yegge reminds me of the cool senior engineer who's young at heart and open to change.
If true to the post, it lacks "real time". Doom scrolling by nature is while chat is async. Refreshing Gmail constantly is not fun.
Edit: also setting up an email interface API to Claude Code seems like a lot more work than just setting up a VPN.
getting a PR back and being able to put comments on it is fine, but ive had middling success getting qcli at least to actually match the comments with the code that was commented on. i get the sense that there isnt any training with the comments inlined well on a diff:/
it doesnt have to be a vpn though, i was on an oauth webbrowser terminal, and things like coder[0] let you run vscode on the browser, including on your phone browser. there's also happy coder[1] which i tried using to connect between the new builtin android linux vm, and skip all the remote stuff entirely, but the phone would inevitably kill the terminal runbing claude, killing the whole thing. you can currently just run claude from your phone in that, which only has the problem that when the vm crashes, all you can do is wipe the partition.
I personally not even convinced that Claude Code any better on average than something like Aider+Gemini 3 or other good model. May be in some specific cases it actually better but in those Aider+'Antropic Model via API' most likely will work too.
with vpn vps if want to interact? how would that work?
“Why not Telegram”
all the crypto bros are already there, and maybe some e-commerce
it could not possibly be easier to get Tailscale up and running on your mac or linux machine, install tmux and mosh on your mac or linux machine, connect to it with Blink Shell https://blink.sh/ on your iOS device that you've also installed tailscale to, and start vibe-coding from anywhere, on a performant, resilient, instantly resumable terminal connection.
seriously, it's a game-changer
This might be the most "when your only tool is a hammer all your problems look like nails" suggestion I've ever read.
Email driven automation isn't a terrible solution to everything - it works very well for support tickets, for example - but it's really lacking in the immediacy required from a serious software development environment.
I'll go further: I think coding on my phone is a fun, neat, idea, and an interesting curiosity, but I don't actually want to do it. There are few situations where I'd feel comfortable getting my phone out to code where I don't also have my laptop with me, and that's going to provide a way better software development experience, so I'm always going to use that for anything serious.
(Sure, there might be small laptops of similar dimensions ... But as the name "laptop" suggests these are made for a different UX... and they require more effort to turn on/off)
I then configured debian-autoshutdown [2] to turn the machine off if there's no traffic on ssh after 15 minutes.
This way I just ssh into my machine (whether via antigravity on my laptop or termius on my phone) and within 30 or so seconds its awake, no physical button presses needed. I documented the whole flow in more detail on my blog [3].
I'm now working on an improvement called machine on proxy (or mop) that will allow me to start Proxmox VMs instead of physical machines, so I can let gemini-cli run wild and if it decides to wipe the entire hard drive I can restore from a snapshot.
[1] https://github.com/simonamdev/ssh-wol-proxy
You don't even have to fully shut down you dev machine, you can allow it to go into stand-by. For that it needs to be wired by cable to LAN, and configured to leave the NIC powered on on stand-by. You can then wake up the device remotely via a WOL magic packet. Maybe this is possible with WLAN too, but I have never tried.
Also, you don't need a Tailscale or other VPN account. You can just use SSH + tunneling, or enable a VPN on your router (and usually enjoy hardware acceleration too!). I happen to have a static IP at home, but you can use a dynamic DNS client on your router to achieve the same effect.
At idle, they aren't loud or consuming much electricity compared to sleep/shutdown.
Fruit co devices in particular are extremely efficient; the Studio is rated at 6W idle, 145W max consumption (cf. https://support.apple.com/en-us/102027 )
My proxy doesn't attempt to handle security. Most folks use either Tailscale or some other VPN solution. In my case I use the wireguard server in my router to VPN into home which gives me access to the proxy and consequently to the machine.
But now instead of swinging back to local resources, apparently we're proposing to add a second layer of remote access (phone -> computer -> Claude servers).
FWIW, I'm not a big fan of AI coding. I use AI (including LLMs) and I am an AI researcher, but the vibe coding just hasn't clicked despite constant efforts. I guess it can make more sense to do it if you're programming from your phone because while normally typing isn't the bottleneck it definitely is on the phone (or at least far less comfortable)
I have an old used Dell Latitude that I use as a pseudo thin client. I ssh into my PC, and everything just works.
I really like this setup because I only have one environment, so everything is there, and I don't have to install anything in the laptop
> I really like this setup because I only have one environment, so everything is there, and I don't have to install anything in the laptop
On that note, one thing I highly recommend to people is to add some visual clues to tell you which machine you're on. I use starship and have a few indicators but I also have some PS1 exports that I've used in the past or use in new tmp instances (I HIGHLY recommend also doing this for when you're using the root account). It can get confusing when you have different tabs on different machines and it is easy to mistake which one you're on.
[0] I also recommend keeping notes there if you like writing in markdown. Files are so tiny that it's worth having them. It's benefited me more times than I can count.
Thanks in advance!
> And there's no way this thing would handle a useful local model
I strongly suggest having a workhorse machine and then let other devices be your terminal into it. Your terminals can be very cheap (or an old machine) or as suggested, your phone.
Nothing could possibly go wrong, those guys are always 100% trustworthy and reliable, contracts and NDAs with them are ironclad and easily enforceable… … o_o
Presumably with an external monitor and the desktop mode it would be better.
Code from tiny llms such as Gemma are a waste of time but it "worked". It was neat to generate a working app completely offline.
The main problem was that the VM crashed on my pixel fairly frequently. Might be better by now.
Edit: found it using these instructions.
https://github.com/nix-community/nixos-avf?tab=readme-ov-fil...
I run htop just fine on my handhelds and I'm pretty sure it sources directly from /proc, /sys or something.
They were fun times :D
You can connect an external keyboard to your phone and if you can swing getting a cheap IPS panel that displays text clearly enough, you’d have a working set up
Anyway, kudos to you, I love reading stories about determination
For anyone looking for a more integrated and smaller approach, I built an open source app builder + runtime: https://github.com/tinykit-studio/tinykit
Basically gives you a Lovable-like app builder with built-in services (database/files/auth/email/payments/etc), content and design fields, and a code editor. Code is a single Svelte 5 file, and you can build/host unlimited apps on one server. And the server is just node + PocketBase, so runs easy on a $2 VPS. And LLM is BYOKey.
What make sense for all users would be a swap-able battery. Water-tightness is no longer and excuse with new phones likes foldables that aren't. Fun fact, Apple dumped the swap-able battery before the iPhone was waterproof.
The example use case images are very funny though! :-)
There aren't that many places in 2025 where getting a phone with internet is significantly cheaper than getting some scrappy laptop or desktop.
No, but it's not a choice between a phone and a laptop. You NEED a phone. So you use what you've got. I've done work helping developers in less developed countries and you frequently find they're sending screenshots of code they've written on phones.
My phone is the original Pixel Fold. You would think I use it unfolded but the passport form factor lends itself to be almost as productive folded that I use it that way most of the time. Unfolded it's just a bit better experience (bigger keys / more display real estate/ more characters per line/ etc).
With that said I'm looking forward to the Click Communicator: https://clicks.tech/communicator
I've also been meaning to write about my setup and open sourcing my tools.
Oh. Writing clojure helps due to the terseness of the language. Not sure it would be a pleasant experience writing something like Java with the 80 character line limit I try to impose on myself
FWIW I do code on the go and I 100% prefer to code at home with my neat setup... but also quite often when I'm on the move and inspiration strikes, I do enjoy having a way to tinker right here and there.
I'll do it only if I have no other choice (i.e. logged into a remote terminal-only server at work). If I have some flash of inspiration I'll write it down in Google Keep and try it out when I get back to my 3-monitor workstation.
But hey, we're all wired different
Fun fact: once you get ssh access to mac, you can control almost anything running on it. Like I added my mac air under termius, and I could mute/unmute any videos playing on chrome using osascript from my iphone :)
I can just ask GitHub to fix something from the mobile app, and then set it to build on PR merge. It works most of the time, but you'd have to be absolutely wacky to do it in production or with any code you actually care about
Essentially you run a server on some machine. Sessions are created in Docker containers, K8s pods, or via Zellij (an app similar to tmux).
You can:
- Directly attach to sessions via Docker attach (built-in via a TUI). You get a normal Claude Code experience, but multiplexed. The switcher/UI shows you the status of Claude and the PR (pushed, merge conflicts, CI status, review status, etc.)
- Manage sessions via a web UI. Connect to Claude Code directly via your browser. You have access to the usual Claude Code terminal or a native chat view.
- Manage sessions via an app. You have access to a native chat view.
It achieves isolation via Git worktrees + a proxy so that containers have access to zero credentials (there aren't even any Claude code creds in the container), which allows you to more safely use bypass all permissions mode.
This works better for me that Claude Code on Web because I have control over the environment Claude is running in. I can give it any Docker image I want, I can have it connect to my local network, etc.
It's still a WIP (the core bits are there, but it's not polished yet), but I'm hoping it provides a friendlier UX with a similar goal for what the OP has in mind.
I'm tired of lugging my laptop around. Let me work from the beach with my phone and ar glasses.
Lately I have observed this algo in myself while learning something new. I constantly code for very short bursts sometimes on the phone or laptop at night, keep jumping between tools and end up consuming more than creating. It comes off as productive but seldom compounds.
A straightforward explanation that has provided me with a helpful point of thought is.
Make a mode selection.
Did conclusions actually occur?
Most doom-coding sessions are loaded with input, no closure.
There are 2 small changes that improved it for me.
Start sessions with a small, visible output goal (one function, one note, one commit).
Time-box input aggressively. I stop scrolling after 15-20 minutes of scrolling.
At the conclusion of every session, I would write what I would do next, even if I don’t do it. ~
Wanting to know how others do this.
Do you intentionally separate learning sessions and building sessions?
Do you have any heuristics to know when you have avoided input?
Or, I thought it was cool until this passage reminded me, "coded a prototype in my downtime" that down time is supposed to be down time.
Life doesn't have down time. Should we avoid learning new things because no one is paying us to learn?
One of my favorite uses of AI is to quickly make some simple 'hello world' level application that I can run using a given technology.
Don't know what an MCP server is? Boot up Kiro and tell it you want to make a sample MCP server and ask it for suggestions on what the MCP server should do. A relatively short while later, with a lot of that time being spent letting AI do it's thing, and you can have an MCP server running on your computer. You have an AI waiting for you to ask questions about why the MCP server does x y or z or how can you get the server to do a, b or c etc
As someone who learns a lot better from doing or seeing vs reading specs, this has been monumentally more efficient than searching the web for a good blog post explaining the concept.
And when I'm doing these learning exercises, I naturally lean towards the domain my company is in because it's easier to visualize how a concept could be implemented into a workflow when I understand the current pain points of that workflow.
I'm not going home and pulling in story's from my board and working on them (generally), I'm teaching myself new concepts in a way that also positions be to contribute better to my employer.
FWIW typically ISP blocks port related to spamming, so usually they block ports related to emails, e.g. SMTP, I believe DNS too, but other ports no problem.
That being said it's quite a silly use case IMHO. If you want to work on a project from "anywhere" then put your project on your server accessible from anywhere, that's literally what servers are for and they cost the price of a coffee per month.
Instead of just jotting down an idea in a notes app (and it sitting there for eternity), I’ll open up Jules, describe the tool, and have it scaffold the HTML. I have Cloudflare Pages hooked up to the repo—once Jules submits the PR, the preview branch builds automatically and I can verify the result on my phone immediately.
I'd say the cycle time largely depends on the complexity of the tools you are building. I've built a movie shelf hooking up with trakt.tv under 30 minutes and a mermaidJS diagram editor spanning multiple sessions and couple of days.
I'd link to a blog post about my setup, but I'm still writing it. Here's someone else's blog post:
No VPN needed.
AI has been changing more rapidly than any other technology I have encountered in my life. It’s absolutely nobody’s fault for not keeping up with it or arriving late to the party, and telling them they should rather just stop because they won’t get it anyway is just awful behaviour.
---
Really, this whole environment of 'coding from my phone with dozens of agents while I'm doing the laundry' feels like satire of the sorts of things we used to laugh at on Linkedin.
To be honest it is already starting to feel that way.
I can build anything with it. Having Claude on top of a terraform repo lets me fully control my infra. Claude is so good at AWS and terraform, and it even found a $3k monthly accidental spend I had running (also sent a refund request to hopefully get some credit back).
Also have a Claude driven CI workflow in GitHub to help keep everything on track.
Having full access to the Claude Code TUI is so much better than the web or iOS interface, plus everything runs on your own setup.
And agree it has replaced doom scrolling / useless new reading.
However, what's the benefit over just using the "Claude Code for Web" feature built into the Claude Code mobile app?
It clones your repo into a VM which has a bunch of dev tools installed, you can install additional packages, set env vars, and then prompt it remotely. The sessions can be continued from the web and desktop apps, and it can even be "teleported" into the terminal app when back at a laptop/desktop.
Would be great to understand what the differences / advantages of OP approach are.
- this is the biggest problem that needs to be solved
- i dont want to keep my computer running 24x7 wasting power for stuff like this
- why not make a robotic arm that you keep at the computer table which can use open cv to plug the computer on when required?
My personal world changed when I discovered Nix On Droid and cloned my personal Claude Code flake which uses pnpm to keep a rolling bleeding edge version with revision controlled dots. I started using Nvim /avante and open router shortly after that, also via Nix on Droid. Game changer for those long subway rides.
If anything, I want to do more of that: get away from the device to let my mind wander. "Doom" coding sounds apt.
I spend hours each week riding transit, and use Claude for a bunch of side projects and have Tailscale set up already, so looks like I’ll be giving this a try this week!
Doom coding might be doomed while I’m in the transbay tube though, with awful cell service…
How’s the diff review? I rely heavily on the vs code integration for nice side by side diffs, so losing that might be a problem unless there’s some way to launch the diffs into a separate diff viewer app on the phone.
Diff review is alright. I'm an amateur programmer. Sometimes I don't look at the code claude generates, but when I'm troubleshooting a bug, I'll ask claude to output all recent changes - which satisfies my untrained eye.
But yeah, you definitely need a native experience to make side by side diffs viable on mobile.
[0] https://github.com/scottbez1/superdiff — I wish I had recorded some videos of the app back then. My code review workflow back then eventually stopped including diff attachments on code review emails, so I abandoned development on it.
And honestly, all this free marketing has me convinced to pay for it
[0]: https://github.com/charmbracelet/crush [1]: https://github.com/baalimago/clai [2]: https://github.com/anomalyco/opencode
I've coded on my phone on several occasions. If you use Android, you don't even need a server or a home computer since Termux works really well as it is. It can run node.js and a bunch of other development tools easily. Or you can just ssh into a server with a development environment and do your stuff their (AI or not).
I run both gemeni (fee) and codex (paid), with tmux thrown in to switch between phone and laptop. Laptop runs vscode with ssh to my server but I could also use the web version of vscode.
- Pydroid
- C# Shell .NET IDE
- Pascal N-IDE
- Shader Editor
It's a fork of Aider but with agent mode, MCP, skills, task manager and more. Very active development team!
Other than that, love it :)
I forgot to say that I _absolutely loved_ the photos!
And thank you! I'm glad you appreciated the humor. I'm still a novice builder, so the thought of ssh-ing to my home computer from a plane geeks me out. I'm about 20 years late but I'm here now!
For me personally I've found two better uses of in-between time:
1. Micro exercises. Really important for health and longevity, especially when it's hard to find dedicated time for exercise.
2. Resting. This means no phone. Yeah hard to resist doom scrolling. Just relaxing muscles and breathing exercises, calming down the nervous system. Increases long term resillience and reduces stress.
So I'm a bit puzzled. If you are in a situation where you can concentrate, why not just pull out a laptop? Typing on phone is really annoying. Even complex conversations with AI I prefer doing on a laptop.
Perhaps there are coding tasks where the prompt is not too complex and it's more about writing code. But you still have to review the result. That's even more annoying on a phone than writing text.
I do however enjoy choosing to do math/coding adjacent activities for leisure or learning sometimes when I'm away from the computer. I've found that it was a net positive in my life to add in puzzles/exercises that I can do with pen and paper in those circumstances.
Lack of space, vibrations etc. even though I can do a lot of work offline if the internet is spotty. It's just not enjoyable.
I prefer to read or chill out.
I kind of envy people who are like oh yeah I coded the feature on the flight... I can't really get in the zone in that environment.
Saying that, I assumed this post was a joke. ssh to a work machine or a personal machine through a VPN is not new, even if you happen to run claude code in that terminal.
I'm interested in these "micro exercises".
Another thing I can recommend is Chinese style radio calisthenics (guang bo ti cao, look it up on Youtube, all Chinese people learn it in primary school and do them daily at school). Full body cardio like and stretching exercises that you can do while staying in one place (you just need space around you). Takes 5-10m, better warming up than just walking and swinging arms and covers a lot of basic things. The entire approach seems virtually unknown in the west.
But also all of the changes I made from the phone were incremental.
Agreed 100% there are healthier uses of my time!
As for too long of a wait I agree, it makes the sessions longer. Ideal window is after a heavy superset where waiting for 3-5 minutes is not a waste.
(Note that I’m not doing this for my real job, just for my personal project)
After you log in you can unlock keychain by running this command
‘security unlock-keychain’
tailscale, tmux, codex/claude code, mosh, blink shell (iOS) https://blink.sh/
i install them via nix-darwin (I've abandoned homebrew)
i am on Tahoe latest beta
their web interface lets you use Claude code and push changes to a GitHub repository
vercel can auto build from a GitHub repo
even less setup and infrastructure needed
(I even use a webcam to capture what my monitor does display when I do remote coding of low level GFX oriented software! Actually my wayland compositor for linux and AMD GPUs)
BTW, IPv6 = ZERO NAT to setup, delicious.
"It's magic".
Yeah what can go wrong when you are travelling and your computer is at home unlocked lmao?
> A Computer running 24/7 with Internet Connection
> A Smartphone
> A Claude Pro subscription
Or.. just install Termux and do it the same way you do it anywhere else?
There’s a few settings I wished were possible, like using volume buttons as modifier keys in Emacs (I’ve heard about this in other apps), but mostly it works fine.
Termux on Android will let you do anything you can do on your standard Linux PC.
I would love to easily be able to set this up easily when a new idea pops into my mind and then have something running (locally or securely in some cloud) within a few hours/days. I wouldn't want to spend a ton of money for this though, nor have a lot of overhead to manage.
Edit: In addition, I'd like some safeguards where I can't have the LLM of choice accidentally delete stuff or do other unintended things on my network.
Otherwise my customers are coders to. they to the same. The difference is the recipient of the order
The loss of the physical keyboard ruined everything for me. I really need the sense of touch.
use a laptop. (trying to do it with only a phone-factor UI is madness.) have a mobile-friendly ISP if desired or needed. solved. been solved for decades
so much of the AI BS hyping is about inventing supposedly unsolved problems. like Google showing me ads to convince me to use Gemini to write a README. no thanks, kids, have been able to do that for many decades using only my brain, eyes, fingers and vi/vim
I’ve actually considered a neck/shoulder support for a laptop in the past but decided against it because it’d be cumbersome and make me a theft target.
As for AI, personally speaking I use AI coding tools to allow me to continue enjoying some hobby side projects with less free time available with a kid. It’s been a massive boost to my happiness in a generally low stakes area. I’m curious to see if I can get a similar unlock on my short and interrupted commute times as well, which is why I (personally) find this article interesting.
Regarding terminals, how often does this requirement occur in practice? Assuming it does, you can probably use your laptop for it, in which case, see above.
Okay, but how are you going to write your AGENTS.md file??