I am a heavy podcast lover.
I spend 30 minutes to 2 hours every day on audio and video — mainly listening to Chinese and English podcasts while running and commuting, and watching long videos on Bilibili while doing strength training. In fact, I started running just so I could listen to podcasts.
Apps on the market are all focused on "discovery" and "recommendation streams," which doesn't quite fit how I consume information. So I thought: why not build one myself? After thinking about it for a while, I finally took action one day. I started by chatting with ChatGPT — talking about ideas, product planning and positioning, and how I should build it. As the direction became clearer, the first version of the plan was basically set.
The product is called ContextEcho, a restrained iOS audio and video playback tool. It deliberately avoids program discovery, social features, recommendation streams, and AI summaries. It only does two things — steady listening + seeing your own accumulation (consecutive days, cumulative time, program rankings, listening heatmaps). Yes, as you can probably tell, I am a data geek.
ContextEcho: Cross-source timeline resume + learning statistics + program rankings
Here are some first-hand observations from me as a builder, which I think might be of some reference value to others.
Observation 1: The bottleneck of personal development in the AI era is not writing code, but whether you "dare to cut"
It took 76 days from the first line of code to the App Store launch, with 468 commits and about 48,000 lines of Swift code, completed entirely in pair programming with AI (Cursor).
But the real difficulty wasn't the output speed — it was that I did three positioning rewrites midway. The most ruthless one involved deleting all the AI features I was proud of at the time (AI annotations, automatic reviews, echo library), a net deletion of nearly 3,000 lines of code at once. The reason was simple: I found that I hardly ever opened those features myself. What I actually used every day was only "listening" and "learning statistics."
AI's default tendency is always to add; subtraction can only be done by the person who actually uses the product every day.
Before deciding to launch, I had already used it myself for 72 consecutive days — build something you can't live without first, then talk about giving it to others. This is my biggest takeaway from these three months.
Observation 2: Traffic will watch your milestones, but it won't verify demand for you
ContextEcho demo video: https://x.com/MaiYangAI/status/2070506319866654834
ContextEcho official launch tweet: https://x.com/MaiYangAI/status/2073088055171571719
Tweet about ContextEcho's ranking after launch:
https://x.com/MaiYangAI/status/2073569387295203710
I posted over a dozen tweets before and after the launch. I asked AI to help me summarize the data, and it's quite interesting to look at:
- Hitting the paid charts: 13.4k views; Official launch: 11.3k; Releasing the demo video: 5.7k
- Daily sharing of "listened for 20 consecutive days": 229; Installing the developer version and inviting beta testers in the comments: 275; Sharing a specific episode I listened to while running: 147
From the data above, you can see what kind of tweets people prefer: milestones and story points get onlookers, while daily usage shares get almost no traction.
Now let's talk about actual conversion, which is how to correctly view the gap between "beta interest on Twitter" and "actual product value." The launch tweet had 11,000 views, but when I publicly invited beta testers earlier, almost no one signed up. I felt a chill in my heart at the time. What really pushed me to launch was the face-to-face recognition and encouragement from a dozen friends at an offline event (Cafe Cursor Shenzhen) — I only bought a developer account after that day. After launching, the app broke into the paid Education category, climbing from 125th at launch to a peak of 23rd.
Looking back, whether to launch or not leads to two completely different paths for a product's evolution: using it only for yourself versus having real users are two different states. When real users in the beta group give direct feedback on problems, the sense of excitement and thrill is very strong — that's the joy of a product truly being pushed out. X can help you spread the story, but real verification comes from putting the product in the hands of users.
To be honest, my app doesn't have many downloads even now, but it is truly helpful and valuable to a small group of people. I think this holds true in the AI era — after all, this is just the first app I've launched. Why set such high demands for myself? Without failure, how can there be success?
Observation 3: An easily overlooked pitfall — ICP filing for the China App Store
Starting in 2023, the China App Store mandatory requires ICP filing. I initially launched without filing, "surviving" for a while during the policy transition period — as a result, when I released the 1.2.0 version update during peak data, it triggered the filing requirement, and the app was directly removed from the China store.
The good news is: individual developers can complete the filing themselves; a corporate entity is not required. I later went through Alibaba Cloud's proxy submission: bought a domain name + a lightweight application server (over 300 RMB, much cheaper for new users), followed the guide to fill in information, answered the audit call, and completed the Ministry of Industry and Information Technology SMS verification within 24 hours. It took about 6 days to get the filing number. After entering it into App Store Connect, the app was back online in about ten minutes. I wrote a complete record of this experience in this article: https://maiyang.me/post/2026-07-14-appstore-china-icp-filing/.
The lesson is just one sentence: If you want to launch in Mainland China, do the ICP filing early. Old apps can survive on the transition period, but as soon as you submit a version update, you are likely to be asked to complete the filing — don't be like me, waiting until the chart data is up and you're excitedly releasing a new version only to be taken down.
Finally, I'd like to hear about your experiences with pitfalls
I wrote these three observations partly to keep a record for myself and partly to start a conversation. Let's talk:
- What is the most painful pitfall you've stepped into? Was it technical, positioning-related, or something completely unexpected like ICP filing?
- Was there a moment that changed your view on "making products"? For me, it was the moment I realized "I don't even open the AI features I was proud of" — it was heart-wrenching, but it also cleared up a lot of things.
Feel free to chat in the comments.
Official website (including screenshots and full introduction): https://www.contextecho.top/
