I Built a Mobile Coding App. What I Actually Use It for Surprised Me.

When I built Kibbler, I had one use case in mind: writing code from my phone. But after using it for a few months, I've realized that "coding" was thinking too small. Claude Code can do a lot more than write functions, and having mobile access to it opens up use cases I never expected.

Checking analytics

I have a bash script that pulls CloudFront metrics and parses access logs to show me landing page stats - page views, conversion funnels, traffic sources, that kind of thing. Running it used to mean opening my laptop.

Now I just ask Kibbler: "Run my landing stats script and summarize the results." I get a clean breakdown of this week's traffic while waiting for coffee. If something looks off, I can ask follow-up questions. "Why did video plays drop?" or "Show me the top referrers from yesterday."

But it goes beyond just running scripts. I can ask Claude to dig deeper into the raw data - "Filter out bot traffic and show me real user sessions" or "What patterns do you see in the conversion funnel?" It's not just fetching numbers, it's having an analyst who can interpret them and suggest what to look at next.

This works for any reporting script you have lying around. Database queries, log analysis, uptime checks - anything you've automated but still need to manually trigger.

Deployments

I used to think deployments required my full attention at a desk. But most of the time, it's just: run the deploy script, watch for errors, maybe roll back if something breaks.

From my phone I can say "Deploy the latest to staging and show me the logs." Claude runs the deployment, streams the output, and highlights anything that looks wrong. If a health check fails, I can ask it to check the error logs or roll back. All without touching my laptop.

This is especially useful for those "I just merged something, let me deploy it before I forget" moments. The change is fresh in my mind, and I can ship it immediately instead of adding it to my mental todo list.

Infrastructure tasks

Checking AWS costs. Rotating secrets. Updating DNS records. Scaling up a service before a traffic spike. These are all things I've done from my phone through Kibbler.

The pattern is usually the same: I describe what I want in plain English, Claude figures out the right CLI commands, and I approve the changes before they run. The approval flow is clutch here - I'm not blindly trusting AI with my infrastructure. I see exactly what's about to happen and can say no.

Last week I set up a new CloudFront distribution entirely from my phone. It took a few back-and-forth prompts to get the config right, but it worked. Would I do this for production-critical infra? Probably not. But for side projects and experiments, it's great.

Project context makes everything easier

One thing I didn't appreciate at first: when you're working within a project, Claude already has context. It knows your file structure, your scripts, your config. So instead of explaining everything from scratch, you can just say what you want.

"Run the deploy script" works because Claude can find it. "Check the error logs" works because it knows where they are. "What's our current AWS bill?" works because it can see your infrastructure config. The prompts are short and natural because the project context fills in the gaps.

This extends to administrative tasks that aren't really "coding" but live inside a project:

• Updating docs - "Add a section about the new API endpoint to the README"
• Dependency management - "Check if any packages have security vulnerabilities"
• Config changes - "Update the staging environment variables for the new feature"
• Writing scripts - "Create a script to backup the database daily"

You get the convenience of conversational prompts with the power of an agent that actually understands your project.

Completely random stuff

And then there's the truly random tasks that have nothing to do with any project:

• Parsing receipts - "Read this PDF receipt and tell me what I can expense"
• Writing emails - "Draft a response to this vendor asking for a discount"
• Research - "Summarize the key differences between these two APIs"
• Quick calculations - "If I invest X at Y% for Z years, what do I get?"
• File management - "Find all screenshots older than 30 days and show me the total size"

Basically anything I'd normally ask ChatGPT, but with the ability to actually run commands and touch files on my machine. It's like having a capable assistant who can do things, not just talk about them.

The mental shift

I used to think of Kibbler as "mobile Claude Code for coding." Now I think of it as "mobile access to a capable agent that can do things on my computer."

The coding part is still the core use case. But once you have that connection set up, you realize the agent can help with way more than you initially expected. Any task that involves your file system, CLI tools, or just having a smart assistant think through a problem - all of that works from your phone.

Try it with something non-code this week. You might be surprised what sticks.