AI Coding Agents

Claude Code for Regular People: Apps, Projects, Local and Mobile

Claude Code sounds like a programmer's tool. In his 15-minute walkthrough, Paul J Lipsky makes a different case: describe a small app in ordinary language, let Code create and run it, then keep the work organized as it grows. His examples are appealing because they solve his annoyances. The important first step for a non-technical reader is choosing a task small enough to inspect and test.

Where to begin

Use regular Claude for an answer or a document. Use Cowork when a multi-step task can be handled in the Claude workspace. Open the Code tab when the result must become a working app or needs local files and commands. Start with one tiny tool, sample data and manual approvals. You do not need to write the code, but you do need to decide what the app may read, change and send.

Watch Paul Build With Claude Code

Source note, 27 October 2026: The timer, Dayboard, stock checker and project examples below are Paul's demonstrations in the supplied transcript. I checked Anthropic's documentation for setup, plan access, local data handling and cross-device sessions. I did not independently build or benchmark Paul's apps. The first-project prompt near the end is an original editorial exercise.

Chat, Cowork or Code?

Paul's central comparison is useful, but “Code does everything better” is his preference, not a rule. Anthropic documents chat and Cowork as one experience during the current rollout, with the exact controls depending on your account. The regular Claude app can create files, too; an app that must be built, run and revised is where Code becomes the more natural choice.

Your taskStart here and check
Explanation, outline or quick draftClaude chat. Review the answer without giving access to a folder.
Multi-step document or connected-app taskCowork. Check which accounts, files and connectors the task will use.
Custom app or computer-based workflowClaude Code. Check its working folder and proposed commands.

For the simplest desktop start, Anthropic recommends installing Claude Desktop, signing in and opening its Code tab. Availability depends on your plan and organization settings. The terminal installer is optional for this guide.

Build One Small App Before Connecting Everything

Paul asks Code to make a menu-bar timer. It writes the app, builds it and launches it in his Mac demo. When he gives Cowork the same request, he says Cowork writes code but leaves the build and launch to him. That is a meaningful workflow difference in this demonstration, not a promise that every platform or app type builds itself without intervention.

  1. Choose one repeated irritation. A single-screen task tracker, timer or reference list is easier to judge than a replacement for your whole workday.
  2. Create a dedicated empty folder. Select it as Code's working location. Keep personal documents, credentials and unrelated projects out of the first test.
  3. Describe the result and its limits. Name the user, three required behaviors, operating system and sample data. Ask Code to explain proposed file edits and installations before taking action.
  4. Run and inspect. Use the desktop preview where available, exercise each behavior yourself, and check where data is stored. Ask for one correction at a time.
  5. Only then add an integration. A calendar or notes connection adds authorization, data-handling and maintenance work. Prove the core interaction first.

Claude Code Desktop's permission modes and diff view let you inspect actions and file changes. Keep approvals narrow for the first build. “It works on my screen” is not yet a tested app; reload it, try an empty state and verify that it starts again after the session ends.

Why Paul's Dayboard Is More Interesting Than a Generic App

Paul's Dayboard brings together his calendar, Apple Reminders and an Obsidian vault. The distinctive part is not the three integrations. It is his own production workflow: copying spoken script lines without stage directions, assigning a video task to himself or an agent, and hiding private information while recording. He also shows a stock-availability notifier, a customized app switcher and games. These are examples of personally useful software, not evidence that AI can maintain any app forever or bypass a store's usage limits.

The transferable method is to write down a friction only you or your team understand, build the smallest interface around it, and test whether it actually saves a step. For a stock checker, respect the retailer's terms and request rate; for an agent-assignment feature, define when a person must review the work.

Code Can Produce More Than Apps

Paul moves from apps to documents, presentations and motion graphics. In one research project, he asks for a report, a deck and an animation, then opens the outputs as artifacts. This is a useful reason to stay in one workspace when those outputs share the same research context. It is not a reason to move every writing task into Code: Claude's regular file-creation workflow can already handle many documents and slides. Choose Code when you need the surrounding files, automation or app build, and review factual claims and visual exports either way.

Use Projects for a Goal, Threads for Separate Jobs

Paul creates an invoice project, attaches an invoices folder, and shows a coordinator dispatching work into threads. A second project develops a report, presentation and motion graphic in parallel. The organizing idea is simple: a Project holds a goal and shared context; a thread handles one job you can open and review. Do not attach a real invoice folder just to explore the feature. Use sample invoices until you have checked folder scope and billing details.

Access caveat: Anthropic says the new coordinator-style Projects are a staged beta for selected Pro and Max Claude Code users. Ordinary Projects have broader availability, but that does not mean every account has the new cloud threads. If yours does not, use one local Code session per small goal and keep the requirements in a file. Paul's interface is a demonstration of his account, not a universal setup screen.

“Local” and “Available on My Phone” Are Different Settings

In a local Code session, the agent works with files and processes on your computer. That does not mean the model runs offline or that no data leaves the device. Anthropic's Claude Code FAQ says source files are read locally and the portions needed for a task are sent to the model API. Scope the working folder and keep secrets out of prompts and readable paths.

Paul also checks a project from his phone. Cloud project threads are designed to continue across devices when that beta is available. A separate local session can appear on your phone only after Remote Control connects it, either per session or through a startup setting. Check the session's environment and Remote Control indicator before assuming it is visible elsewhere. Local work and cross-device access are separate choices.

A First-Project Brief You Can Copy

This original prompt is for testing one internal-tool idea without real customer data or accounts. It asks Code to clarify the job, build a small local version, and show what it actually verified. Review every proposed action before approving it.

First-project brief

Test one small internal tool

Sample data, narrow permissions and a real check.

Ready to copy

Video Chapters

Get oriented

TimeTopic
00:00Why Paul made this guide
00:33Claude and Claude Code

Build and create

TimeTopic
01:39Build a small app
03:03Dayboard and custom app examples
06:11Documents, presentations and motion

Organize and access

TimeTopic
07:32Projects and threads
13:15Work in a local session
13:54Continue from a phone
14:42Final thoughts

Sources and Access Notes

Product screens, plan access and settings can change. Check the linked Anthropic documentation and your own account before following a screen-specific step.

Common questions

Do I need to know how to code to use Claude Code?
No. You can describe an app in ordinary language and review the result. You still need to check permissions, test the workflow and understand what data an app reads or changes.
Are local Claude Code sessions fully private or offline?
No. Local means the agent works with files and processes on your computer. Anthropic says task-relevant portions of files are sent to the model API; a local session is not offline inference.
Can I open every desktop session on my phone?
No. Cloud project threads can be available across devices when your account has access. A local session appears on other devices only when Remote Control is connected for that session or enabled at startup.
Does everyone have the new coordinator-style Projects?
No. Anthropic describes that experience as a staged beta for selected Pro and Max Claude Code users. Ordinary Claude Projects have wider availability but do not imply the new parallel-thread behavior.
Share
X LinkedIn Reddit
Build Yours

Want a system
like this one?

Book a free 30-minute call. We map your situation, identify the highest-impact automation, and figure out if we are a fit.

Book Free 30-min Call