Vibe Coding: how I (try to) get smarter alongside AI -- Building a Knowledge Tree
Vibe coding is real.
You can ship a working app in minutes.
What’s harder is knowing what you actually understand.
This is the 3-part workflow I use to keep learning while I vibe-code.
- Critical Thinking: parallel window for questions
When I vibe code, I keep a second window open. That’s where I ask my coding agent to explain things as we go, especially when something feels non-obvious.
Example: yesterday while installing @openclaw , I noticed it was using bun.
That triggered a quick detour:
• what bun is
• how they work
• why bun instead of npm
Nothing fancy. Just slowing down enough to ask why instead of blindly accepting output.
- Memory: build your knowledge tree
One thing I noticed early on: good explanations disappear into chat history way too easily.
To avoid that, I connected my coding agent (@claudeai) to a memory layer (@ensue_ai ).
Whenever the agent explains a concept, gives an overview, or shares a mental model, I ask it to save it into a structured personal knowledge base.
Over time, this turns into something like a knowledge tree - things build on each other.
I automated this by forking the ensue-memory-network skill and creating a new Claude skill called learning-memory, which saves what I learn using a consistent structure.
- Recall: make things stick
Perhaps it's just me, but even when something clicks in the moment, I tend to forget it if I don’t revisit it.
What’s helped is baking recall into the workflow:
• at the end of a session, I ask the agent to summarize what I learned
• when I start the next session, it looks at that memory and quizzes me on it
It refreshes my own context fast and makes it easier to keep building on previous knowledge, putting me in a much better place to continue.
The learning-memory skill has hooks that automates this.
Why this feels useful
I’ve started using this approach recently, and it’s already paying off.
I still get the speed and leverage of vibe coding. But over time, I feel more confident reasoning about:
• what the code is doing
• why certain tools or patterns are used
• what tradeoffs I’m actually making
Speed doesn’t have to kill understanding.
With the right loop, even vibe coding can develop taste.
If You Want To Try It Out: setup
You could probably implement something similar with other tools, but this is my current setup:
• AI agent: Claude Code
• Memory skill: enter "/plugin christinetyip/ensue-learning-memory" in Claude to install the skill
• Memory layer: Ensue (you can generate a free API key at ensue-network(dot)ai after signing up)
• Last step: enjoy vibe coding and growing your knowledge tree.
Author's self-reply in comments:
Added a script in the repo to visualize the knowledge tree: https://t.co/T6vpeBMj4k
@leeparayno Yes, you can store all that context on Ensue and retrieve it later, including decision traces and relationships, e.g., *why* features were built, so you can apply the same decision logic to future features.
Check out https://t.co/YoEumvTjp1 if interested or DM me if you get
@ab_edge Cool, how did you implement your system? I've been thinking about this a lot, and I'm very curious how others approach this
@jasonxsigmon Yeah, having a personal reference you can revisit is very helpful.
I probably went a bit overboard turning it into a skill, but it was a lot of fun, and it’s made me really motivated to keep growing the knowledge tree.
How are you structuring your notes and references as it