"Independent Development Diary 6: This is Not a Delay, But I Want AI to Truly 'Understand You'"

Personal Life

This week has been quite poor; every time I return to the country, I end up suffering from rhinitis and then cough for a month. The first two times, I took medication for a month, but this time I didn't go to get a prescription and plan to tough it out. Because I previously had a fungal lung infection, I am very fearful of prolonged coughing, and my progress has slowed down a lot recently. In fact, many times, working on an independent product is just like having a job; if you stop working, you stop progressing, because you need continuous output and exposure for people to see your product.

A friend visited this week, and we went to an outer province to eat durian. I remember when I was younger, I didn't eat durian; I thought it tasted terrible at that time. However, after coming to Thailand, I found many high-quality durians, and after trying a sample, I discovered that I could accept it and it was quite tasty. I won't write too much this week; I feel I need to rest more.

Last week's post unexpectedly gained some traction, reaching 10,000 views, but it didn't bring much engagement. I also received some negative feedback, but I've gotten used to it. Since the previous release of StickerAI, I received a lot of criticism, some of it harsher, and I still can't understand why there is so much negativity online. A friend told me that no matter what you do, there will be people who like it and people who don't; what you need to do is block those who don't like you. Of course, some say that this will lead to an information bubble. From my observation, most of the negative comments come from users who haven't used your product or who start criticizing after just reading a couple of words. If someone is a paying user and criticizes your product for specific shortcomings, you should listen; if they criticize while still using it, it indicates there is a genuine need.

Regardless, it's important to distinguish between genuine product needs and mere venting of opinions. I won't write too much; I'll briefly discuss product progress.

StickerAI

This week, I made a small optimization. I had been too fixated on optimizing the AI's prompts to solve the display issue of images in dark mode. I tried several versions but was not satisfied. One day, I suddenly thought that since our images have a transparent background, could we differentiate between the image itself and the transparency? After discussing with the AI and researching some materials, I came up with a solution: after generating the image, we can apply an additional process to add a white outline around the non-transparent edges. I'm not sure if you can see the difference below.

PromptPlan

After releasing the first version, we registered fewer than 50 users. Since it allows usage without logging in, the exact number of users is unknown, and we didn't add internal statistics, so we are unclear about how many people are using it.

The first version of the MVP basically met expectations. I originally planned to release a new version this week, but due to health reasons and a friend's visit, progress was delayed. Most importantly, the difficulty of the new features exceeded my expectations, so it is still in debugging.

The new version will likely add the following three features:

Some friends may not understand what MCP is, so I won't go into detail; simply put, it allows the AI to access other services on your behalf. For example, if you want to access files in a system folder, the traditional approach is to upload the files to the AI and then ask. With MCP configured, the AI can check this content itself and return the results you need.

Many friends may still be unclear, so I'll describe a few current scenarios. We are currently doing some intelligent optimizations based on user background.

Scenario 1: User has not filled out background information

User input: 【Please help me plan the articles I need to write this week】

Result: 【This week, write an article on this task】

Reason: Due to the lack of context, the AI may only convert limited information into an independent task.

Scenario 2: User has filled out background information

User background information: I am an independent developer, and I regularly publish an update log article about StickerAI and PromptPlan every week.

User input: 【Please help me plan the articles I need to write this week】

Result: 【This week, write an article on the new features of StickerAI】 and 【This week, write an article on the new features of PromptPlan】 (2 tasks)

Reason: The AI knows you are doing these things, and after analyzing your intent and background, it provides these two tasks as results.

Scenario 3: User has filled out background information and enabled the file system (MCP)

User background information: I am an independent developer, and I regularly publish an update log article about StickerAI and PromptPlan every week.

MCP: File system service

User input: 【Please help me plan the articles I need to write this week】

Result: 【This week, write an article on the new features of StickerAI (description shows added white outline, fixed xx bug, etc.)】 and 【This week, write an article on the new features of PromptPlan (description shows added MCP feature)】 (2 tasks)

Reason: By knowing the background information, we understand that we are working on the update log articles for StickerAI and PromptPlan. Then, through the file system service, we can read our code, requirements, and development logs to know exactly what we have done, and then convert that into notes written into the tasks.

Of course, the above is my rough idea, as the content returned by different large models may vary, and I don't have a particularly standard answer. However, MCP allows our AI to access a multitude of possibilities; for example, it can search the internet for trending content and convert it into tasks.

Currently, the configuration for reading has been mostly completed, but combining it with AI planning has been under debugging and modification, and we haven't been able to achieve the above effects, which is one of the reasons for our delay this week.

Additionally, once the new version is released, the original price will be restored. The current price was set as a pre-sale expectation; after all, I believe most people are not very willing to pay for an import-export feature. After this version is released, PromptPlan can truly be called an AI intelligent planner. That's all for today.