This article was written by Typeless in memory of that magic wand I just found. Over the last month, I have noticed a great leap in my usage of AI. It all started with Claude Code.

Because Augment Code increased their price, my pro plan was no longer enough and I couldn't renew it to cover my monthly development needs. This forced me to get rid of my laziness and try to configure Claude Code.

Initially, I met a lot of obstacles. For example:

1. The MCP login was constantly unauthorized.
2. The $100 or $200 price point was intimidating compared to a $15 "sweet spot" plan.
3. The pro plans had very limited session usage that reached the gap every few hours.
4. The TUI terminal was hard to use and not aesthetically pleasing.

However, after I finally set up the VS Code extensions and enabled the "bypass permissions" YOLO mode, new gates opened for me.

High-Intensity Development

I developed with Claude Code and finally understood its difference from Augment Code. You need to use high intensity in "plan mode" and utilize magical prompts to let it generate or compile several ranked plans. While its out-of-the-box experience is far behind Augment Code, proper configuration allows it to connect to:

• To-do lists
• Calendars
• MCP (Model Context Protocol)

It uses various skills to manipulate very large codebases and can run independent loops for a long time. It also supports multi-threading and concurrency. I noticed you can use the sidebar, open VS Code, and run several threads of Claude Code in command with Featurebase and Linear MCP to consume all feature requirements, bug fixes, and tests very efficiently.

Remote Control via HappyCoder

The last thing is remote control by HappyCoder. This is the next amazing, unbelievable, fabulous tool I've found. A long time ago, I tried to use SSH and a terminal to access a remote VPS server from my mobile, but it had an ugly appearance and interacting with a mobile terminal was very uncomfortable.

With HappyCoder, my old MacBook Pro (which had been asleep for a long time) has turned from waste to gold. My typical workflow now looks like this:

1. Architect Design: (a) Write HLD (High-Level Design) documents. (b) Generate a product vision in Markdown. (c) Spend two or three days writing more than 5,000 words of detailed requirements (must-haves and should-haves).
2. Validation: Run Claude Code repeatedly to check if the HLD is missing important features or has self-conflicts, clearly splitting features from V1 to V2.
3. Specification: Once there are no conflicts, I turn the HLD into more than ten SPEC documents.
4. TDD System: Design a Test-Driven Development system including unit tests, SQL, and E2E (End-to-End) tests.
5. Implementation: Build the code until it passes hundreds of test cases. The code itself will manage the process of dependencies and finally output usable code.

Currently, I can use HappyCoder from my bed 24 hours a day. Even when I am in a restaurant eating, I can control my AI agent at home running on the MacBook Pro. I can even let it:

• Control iOS or Android emulators.
• Use MCP to communicate, take screenshots, and tap buttons for the next steps.
• Pack and build the APK.
• Upload it through Catbox for system sharing so I can download it directly to my phone and provide feedback.

This is such an amazing feeling. This is the next generation of Siri—one that actually only requires me to speak, and it will finish everything.

In the end, here are some supplementary details:

1. Remote control and tool limitations: (a) Claude Code has limited ability as a GUI package for certain protocols (like MCP authentication). (b) For example, if you need to check cloud usage or handle MCP authentication that requires typing a password in the terminal, it cannot deal with that directly. (c) In these situations, I use Tailscale to quickly create a virtual VPN so my mobile phone terminal can connect to my MacBook Pro as a server anytime using SSH 2.0. This is a simple but necessary workaround for permission-related issues.

2. Tool usage across different environments: (a) For simple tasks, such as to-do format requirements or calendar queries, using cloud-based tools is sufficient. (b) For complex engineering development—where you might have several plans that need to be evaluated, ranked, combined, or slightly changed before sending—I use a "magic app" I developed myself called PromptNote.

3. Features of PromptNote: (a) I use it to create "Snips" (versioned snippets) where you can like, unlike, or favorite different versions you've tested. (b) This is useful for various prompt engineering situations, such as asking for client needs, searching the web for comparison solutions, finding the best prompts, debugging, or writing HLDs and specs.

I am currently using this "super workflow" to develop PromptNotes into an input method for mobile phones. I am simultaneously developing an iOS keyboard extension and an Android IME that will be bundled with the main application.

The user experience is very smooth:

1. You just need to press a double slash (//) or click a single button.
2. This brings up the Snips button, allowing you to quickly search, filter, and click to send the "magic comments" you need.

One last thing I have to mention is that Typeless has finally closed the loop. Before, you had to type everything on a mobile phone or laptop, which is slow and requires a lot of cognitive effort. But with the help of Typeless, you can just speak and output all of the thinking and thoughts in your mind, quickly expressing production requirements. This allows you to quickly express and capture production requirements, and create new tabs and transition to new requirements faster.

It instantly clears the bottleneck within the entire loop. I am becoming a "super-node" (the intermission point of the whole information network). Now that the last mile of input and output has been smoothed out, I am fulfilled with the desire to express many thoughts that could not be written down before. Now, I can capture them all. I am currently waiting for Typeless to release their Android version; I am already quite eager for that.

Finally, here is a "gold sentence" quote from myself:

Damn, I suddenly realized a profound truth: social status and influence are inversely proportional to screen/input device size.

At the bottom end of the service industry, you’re standing at a counter all day. A step up, you’re in a security or control room staring at ten thousand giant monitors. Average folks are in office cubicles looking at a few screens. Big shots are in meeting rooms or business class looking at small laptops or TV polls. The heavy hitters—CEOs and Academicians—only need a foldable phone in first class to command the world. The ultimate leaders have human secretaries to listen to what you say; they don’t use screens at all.

Now with Claude Code, you only need to give voice commands to your phone. Combined with Test-Driven Development (TDD), you can completely break free from the screen and focus on real life. Go out and have fun, periodically issue a command, and push to production with one click to test directly on your phone.

And:

AI is working overtime non-stop to create tools (referring to PromptNotes) that allow humans to supervise it so it can work overtime even better. It’s giving off major “DingTalk development team employee” vibes. LOL.