muhammad.dev

Tag: productivity

  • My 2026 Development Workflow: Claude Code Desktop + Cursor

    My 2026 Development Workflow: Claude Code Desktop + Cursor

    I’ve been refining my development workflow over the past few months, and I’ve landed on a combination that feels incredibly productive: Claude Code from the Desktop app paired with Cursor for code review. Let me walk you through how I work.

    Anthropic recently added a native “Code” tab directly inside the Claude Desktop app. It is a GUI for the engine that powers the CLI. One notable feature is its support for isolated Git worktrees. This means you can have a brainstorming conversation in one tab while a “Code” session runs in another, making changes in a separate worktree that won’t touch your main working directory until you’re ready to merge.

    Why not git worktree? My workflow is simple enough that I just work directly in my main/feature branch and review everything in Cursor before committing. But if you’re juggling multiple experimental features or want that extra safety net, the worktree support is there.

    Why not the CLI? I know many developers swear by the Claude Code CLI, but I’ve found the Desktop app suits my workflow better. There’s something about having a dedicated window for my AI conversations that keeps things organized. I can easily reference previous discussions, and the interface feels more natural for the kind of back-and-forth dialogue I have when working through complex problems.

    The Workflow

    Here’s how my typical development session looks:

    1. Brainstorm with Claude Desktop

    Before writing any code, I start by talking through the problem with Claude. I describe what I’m trying to accomplish, share relevant context about my project, and bounce ideas back and forth. This conversation helps me clarify my thinking and often surfaces edge cases I hadn’t considered. It’s like having a patient colleague who’s always available to rubber duck with.

    I then ask Claude to generate a prompt for Claude Code based on this discussion.

    2. Generate Code with Claude Code

    I open Claude Code in the Desktop app and paste in the prompt generated from step 1. It would have context about my project structure, the technologies I’m using, and any constraints I’m working with. Claude generates code, explains its reasoning, and I can ask follow-up questions right there in the conversation.

    3. Review in Cursor

    Once Claude has generated the code, I switch to Cursor. This is where I put on my reviewer hat. I don’t just blindly accept what the AI produces—I read through it carefully, understand what it’s doing, and verify it aligns with my project’s patterns and standards.

    Cursor’s diff view makes this review process smooth. I can see exactly what’s being added or changed, accept individual hunks, or modify the suggestions before committing.

    4. Test, Accept, and Commit

    After reviewing and testing, I accept the changes I’m happy with and commit them with meaningful commit messages.

    Why This Combination Works (for me)

    The separation of concerns is what makes this workflow powerful:

    • Claude Desktop for Brainstorming — This is where ideas take shape. I describe the problem I’m solving, share context about my project architecture, and have a back-and-forth conversation to explore different approaches. Claude helps me think through edge cases, consider alternative implementations, and refine my requirements before writing any code. By the end of this phase, I have a clear mental model and a well-crafted prompt ready for code generation.
    • Claude Code Desktop for Generation — With the refined prompt from brainstorming, I switch to Claude Code which has direct access to my codebase. It understands my project structure, existing patterns, and dependencies. The code it generates is contextually aware—it follows my naming conventions, integrates with existing modules, and respects the architectural decisions already in place. I can iterate here too, asking for adjustments or alternative approaches.
    • Cursor for Review — This is my quality gate. I examine every diff carefully, understanding not just what changed but why. Cursor’s interface makes it easy to accept good changes, reject problematic ones, and make surgical edits where needed. This deliberate review process ensures I never ship code I don’t understand. It’s also a learning opportunity—I often discover new patterns or techniques by studying what the AI produced.

    This two-step process forces me to slow down and actually review what the AI produces. It’s easy to fall into the trap of accepting AI-generated code without understanding it. By deliberately switching tools for the review phase, I create a mental checkpoint that keeps me engaged with the code.

    Tips for This Workflow

    1. Be specific with Claude — The better your prompts, the less cleanup you’ll need in Cursor
    2. Review as much as possible — Don’t let the convenience of AI make you lazy about code review
    3. Commit incrementally — Small, atomic commits make it easier to track what the AI contributed
    4. Keep learning — Use the review phase as an opportunity to understand patterns you might not have written yourself

    Final Thoughts

    AI coding assistants are powerful tools, but they work best when you stay in the driver’s seat. My Claude Code Desktop + Cursor workflow keeps me productive while ensuring I remain the decision-maker for every line of code that ships.

    If you’ve been looking for a way to integrate AI into your development process without losing control, give this approach a try. The key is finding the right balance between leveraging AI’s capabilities and maintaining your own understanding of the codebase.

    Thanks for reading. Let me know whether you agree/disagree or have a different take.

  • Siri-controlled DND 🎧 for my office room

    Siri-controlled DND 🎧 for my office room

    Pausing notifications on your phone is great. The iPhone’s Do Not Disturb (DND) feature is also great. Would it not be awesome if we had a DND for our room/office? Well, that was when I had an idea for a mini project.

    My nephews and I went to a local electronics store and bought some gadgets. I just bought the ESP32-C3, which has two USB-C ports.

    These were the steps I did in order. Of course, ChatGPT helped me every step of the way.

    1. I ensured the ESP32 was working by running a blinking sketch, which I created using Arduino IDE.
    2. I then created an on-off switch using the BOOT button on the ESP32. This would turn the LED to switch between Green and Red alternatively each time you press the BOOT button the the microcontroller.
    3. Then I connected the ESP32 to Blynk.
    4. Once it was connected to Blynk with the token, etc. I used three links:
      • One would turn the Green light on when a GET request is sent to it.
      • The other would turn the Red light on when a request was sent to it.
      • The third one would give the current status of the light.
    5. Then I did some configurations using Siri Shortcuts. They were:
      • Office Green: which would essentially visit the above link I spoke about followed by turning off Do Not Disturb on the iPhone/Apple Watch.
      • Office Red: which would essentially visit the second link above, followed by turning on Do Not Disturb on the iPhone/Apple Watch.
      • Office Switch: this would get the current status of the light. If the currentStatus was 1, it would run Office Green, otherwise it would run Office Red.

    The end result was something like this:

    In case you have any inspiring ideas for such projects, feel free to build on it and share to the world!

    If you find this post inspiring, I have done my job. If you asked yourself why not just close the door instead of this contraption? Well, then you have done your job of critical thinking.

    Cheers!

  • Falling in love with boredom

    Falling in love with boredom

    I had a phone call yesterday with someone I admire. We were speaking about various matters. Then the discussion came down to my career and how I got to be where I am. We were discussing the significance of a college degree to a successful career in software engineering.

    (more…)
  • Social media as a means to an end

    Social media as a means to an end

    This blog post is about using social media mindfully, from the perspective of a software developer who has benefited from it. Read on if you feel this will be useful to you.

    (more…)
  • On Working Remote

    On Working Remote

    I have been working remotely for over a year now. I am completing one year at rtCamp soon and before that, I was doing some consulting remotely. During this period, I thought about how I could make remote-work work, and this meant I read some articles and books from people much smarter than I am.

    (more…)
  • Evernote or Keep – or both?

    Evernote or Keep – or both?

    One of the first productivity apps I used on a smartphone was Evernote. Needless to say it blew my mind away – the simplicity, ease of use and the sophistication too.

    Then came along a product by Google called Keep. I tried this and never looked back – except when I needed to check for older notes on Evernote.

    (more…)