
Staff Software Engineer 🧑💻 | Applied AI 🤖 | ex-Grab/MyTeksi 🚖 | @Nuriwa_J 💕 💍 | writes 👉 https://xitter.cf/t.co/55qlXX3LaE
Malaysia / Singapore
Joined December 2007
- Tweets23.3K
- Following1.4K
- Followers1.5K
- Likes19.3K
Pinned Tweet
Imagine if there is a social media protocol,
instead of a social media platform.
Imagine if you could send,
Notes and
Other
Stuff
Transmitted by
Relays
npub1pq3nweafwtwelmzmz2s9w0uafd68jm803m959r9wfghkhqdky3ys9feyc5
Yes, I do feel the same way, it is actually more mentally exhausting working with AI than before.
I am working on approximately 17 things simultaneously, and frankly, I find it overwhelming, exhausting, and not very rewarding.
I used to find extreme joy in living with a problem. Keeping all the details in memory, thinking about it deeply for days, iterating and improving.
Now all of that is gone. I am still pretty good at guiding agents - decades of developing and shipping all kinds of software cannot be erased - but I feel that what I considered one of my biggest strengths is diminshing in value.
Anyone working seriously with agents knows this, even if they don't want to admit it. The deep super focused thinking is much more rare now. It is like a muscle - if you don't train it regularly, it gets weaker.
Is that something that we are willing to give up? Something that is no longer needed? Can the new generations of engineers do without it? No idea.
btw I kinda build driftradio.net for fun 😁
Personal take: The games in Alice in Borderland season 3 was good, really good, but the ending on season 2 was a great way to end and sum up everything. We don't really need season 3.
I've been thinking...
Highly skilled and very experienced coders who never really enjoyed the typing part of programming are the ones that are going to thrive in this AI-agentic future.
More stations. 20 Singapore based stations.
driftradio.space
We just added 20 more stations from Singapore based.
Don't scroll, drift.
driftradio.space
Hear me out ... the folks that made the Rurouni Kenshin live action movies should make Samurai Champloo live action.
Yesterday I added 59 new stations mostly from Malaysia. Never thought there were so many station still broadcasts in SHOUTcast / Icecast
driftradio.space
The dial just got a lot bigger. 59 new stations on The Space from Malaysia and Singapore —> tune in at driftradio.space
Lately, I’ve been thinking about what does “source of truth” in a coding project means to me. Is code the source of truth?
I don’t think so. Code only tells what is implemented. It does not tell you what should be implemented. Intent is the missing layer, and intent is the source of truth.
Tests used to be the place where I codify my intentions. I should do this, it should not do that.
When I change something and if an unexpected test fails, it gives me a signal to either reevaluate the intentions or the change I’ve made.
AI coding agent breaks this friction. When a test fails, the agent often edits the test to get the pipeline green. The intent-encoding layer gets silently rewritten.
I can try to restore the signal directly by telling the agents not to modify existing tests, but this would just slow things down.
Lately, I’ve been keeping a Karpathy-style LLM wiki inside the codebase itself. Committed and version-controlled. This is where I write intentions and decisions made in natural language.
“Feature should not do this, it should only do that, so it aligns with this-and-that.”
When I have my agents review the changes it made, I ask it to check with this wiki notes. If the changes go against something that was decided, I tell it to highlight it to me. I’ll decide to either change the notes or fix the changes.
Sometimes a decision was made while in planning mode with the coding agent, I just tell it to update or add a new note to the wiki folder about what we decided.
You can say that this sounds like a PRD document. But it is more organic and fluid as it grows and evolves as the product grows and evolves.
In fact, if you have a PRD, that should be the source of truth, and it should sit very close, preferably next to the code itself, and AI agents making changes should check against it.
I know documentation can drift, especially when humans are maintaining it. You push an emergency update one time on a late night and forget to update the docs. It becomes stale. AI agents are actually better software engineers than us humans, in the sense that they are better at keeping docs updated.
Tests were always an imperfect proxy for intent, and code only shows what was built. So wherever you keep the intent, it must be co-located, versioned, and readable by AI agents.
AI does not make us mediocre. We were already mediocre.
AI does not make us stop thinking. We had already stopped thinking. We were already parroting whatever we saw, whatever the algorithm put in front of us.
I agree that we need to find our own voice, but it is not AI that is stopping us from having our own voice, we just never bothered to find our own voice.
If you use AI for everything, this video is for you.
#ArtificialIntelligence #AItools #Claudecode #ChatGPTpromprs
Announcing Drift Radio 📻
I’ve always wanted to do something like this. To play around with SHOUTcast / Icecast streaming Internet radio interface. And implement an IRC-like chat protocol.
It’s now live give it a try.
Drift Radio is:
❌ not a start-up
❌ not a SaaS
❌ not a social media platform
It is a place to drift while listening to Internet Radio. I’m there right now listening to one of the stations. Come find me.
Don’t scroll, Drift.
driftradio.net
Drift Protocol is live. Drifting into a station now creates a temporal chat space. No account. No profile. No archive. Just connect.
driftradio.net
I'll be here for the next 20-30 minutes or so..
driftradio.space/deephouse-f…












