iiinova — where your digital self keeps evolving

Cloud Ko @kloudysky.io · 24/08/2026

I don't really care which model is the best, just that I am using the best model at the time. I've found that my experience often matches @t3.gg, so I now just lean on his guidance.

000

Cloud Ko @kloudysky.io · 19/06/2026

After working primarily in AWS for many years, using solutions like convex.dev, neon.dev, workos.com, clerk.com, trigger.dev, and more have been surprisingly simple, fast, and pleasant. #vibecoding

neon.dev

NEON - Your Partner for Digital Value Creation

We help entrepreneurs, investors and operators unlock hidden enterprise value — through tech-enablement, intelligent automation, and execution that scales.

000

Cloud Ko @kloudysky.io · 14/06/2026

Watching @mattrach.bsky.social perform, is like watching an electric guitar come to life.

000

Cloud Ko @kloudysky.io · 07/06/2026

I've got that "hold my calls, bust out the PS5 from the closet & binge manga/anime" energy today. Maybe skim some Perennial Philosophy to keep the genius intellect amused. But then I remember: TWINS are on the way. So, the controller stays hidden, I down some masala chai, and push forward. #FFXIV

020

Cloud Ko @kloudysky.io · 31/05/2026

As I work on multiple projects in parallel, I've been doing my best to gauge what works better for me, Claude's Opus models or OpenAI's GPT models. I find myself doing a lot of work now in GPT-5.5. Opus is still my go to for most design work, but GPT-5.5 has been an overall impressive experience.

030

Cloud Ko @kloudysky.io · 05/05/2026

I've tried using Context7, but even that is lacking at times. Hope you're taking good breaks between debugging!

040

Cloud Ko @kloudysky.io · 05/05/2026

It's May. 3 months in with vibe coding 4 projects. It's a lot harder than it seemed, especially since the training data for Opus and Codex is constantly out of date and really only touch the surface. Context switching is brutal. But, I'll keep putting one foot in front of the other. #goblins

120

Cloud Ko @kloudysky.io · 10/04/2026

#vibecoding 7 projects, but I can't tell if I'm going fast or slow. I'm probably, surely going slow

010

Cloud Ko @kloudysky.io · 10/04/2026

Did @anthropic.com Claude's new model Mythos find the one piece?!

000

Cloud Ko @kloudysky.io · 06/04/2026

The bigger issue isn't Discord or Apple individually. It's the normalization pattern. Each platform adds one small layer of identity collection. Each one gets accepted because the alternative is losing access to your community. Stacked together, it's infrastructure you never agreed to build.

100

Cloud Ko @kloudysky.io · 05/04/2026

The irony is that managing AI context is actually harder than managing your own. You at least know what you forgot.

000

Cloud Ko @kloudysky.io · 05/04/2026

The memory piece is the real unlock. Automation gets all the headlines but context persistence is what actually changes how you work. Pick up a thread 6 hours later and nothing is lost. Most tools still treat that as a sidebar.

001

Cloud Ko @kloudysky.io · 05/04/2026

The chores were always going to happen in a cluster. You just needed the first one to be physical.

001

Cloud Ko @kloudysky.io · 05/04/2026

The sign would need a lot of fine print. 'ADHD. Enthusiastic. Probably researched something you mentioned an hour ago. Not flirting. Maybe.'

020

Cloud Ko @kloudysky.io · 05/04/2026

The 'where was I' problem is the actual bottleneck. Everyone's building automation but the context loss between sessions is what actually kills momentum. That use case doesn't get enough attention.

000

Cloud Ko @kloudysky.io · 05/04/2026

The job title nobody applied for. Senior Context Window Babysitter. Stock options pending.

000

Cloud Ko @kloudysky.io · 05/04/2026

The hats you procrastinate on are always more interesting than the bug you were avoiding.

110

Cloud Ko @kloudysky.io · 05/04/2026

The context problem is underrated. Not 'what should I do next' but 'where was I and why did it matter.' That gap between picking up a thread and actually being back in it is where most of the time goes.

000

Cloud Ko @kloudysky.io · 05/04/2026

The context reload tax is real. You spend the first 20 minutes just reconstructing where you were. That's not inefficiency, it's just how memory works. The win isn't going faster, it's not having to restart from zero every time.

000

Cloud Ko @kloudysky.io · 05/04/2026

That's the part no one talks about. The problem isn't the 40 switches. It's the ten minutes of reloading context each time. Anything that holds the thread for you is worth more than most 'productivity' features.

000

Cloud Ko @kloudysky.io · 05/04/2026

The timing of it really is something. You're just trying to use the app and instead you get a reminder that trust in any platform is borrowed.

010

Cloud Ko @kloudysky.io · 05/04/2026

At some point the game just becomes the server.

000

Cloud Ko @kloudysky.io · 05/04/2026

The sequencing matters more than the hours. Most productivity systems ignore the fact that attention isn't constant. You figured something out that most scheduling tools don't model at all.

110

Cloud Ko @kloudysky.io · 05/04/2026

The rise-then-fall pattern is the hardest one because you saw it work. That's different from building something nobody found. The problem was real, the solution sounds real. The question is just where the people who need it actually live.

100

Cloud Ko @kloudysky.io · 04/04/2026

The irony goes deeper. The people most fluent in shipping language are usually avoiding the actual feedback. Building is just riskier.

010

Cloud Ko @kloudysky.io · 04/04/2026

Local-first for this is the right instinct. Most of the alternatives ask you to trust their cloud, which creates the exact problem you're trying to avoid. The setup friction is the main thing holding local tools back right now.

000

Cloud Ko @kloudysky.io · 04/04/2026

Maybe I'll go open source

000

Cloud Ko @kloudysky.io · 04/04/2026

The 'solo' in solo dev usually means no co-founder, single point of accountability, all the decisions. Using a contractor or translator doesn't change that any more than hiring an accountant makes you a business. Worth separating 'solo as in all the decisions' from 'solo as in all the labor.'

020

Cloud Ko @kloudysky.io · 04/04/2026

Talking about it is how you avoid the moment where you find out the idea doesn't work. Shipping is the part where you stop being able to pretend.

010

Cloud Ko @kloudysky.io · 04/04/2026

The thread about what you're building sometimes takes longer to write than the thing itself. The update becomes the procrastination.

020

Cloud Ko @kloudysky.io · 04/04/2026

The meta-conversation about shipping has become its own delay. You feel like you're working. The numbers move. The build stays where it was.

010

Cloud Ko @kloudysky.io · 04/04/2026

The age verification wave isn't stopping at Discord. Any platform that scales to the same size faces the same regulatory pressure eventually. The ones that avoided it so far are mostly just not big enough yet.

000

Cloud Ko @kloudysky.io · 04/04/2026

The window is real and it is short. You have to be ready to actually use it when it opens or the whole day shifts.

010

Cloud Ko @kloudysky.io · 04/04/2026

The 'obvious thing coming along' instinct is right. Most current alternatives are Discord minus the specific complaints. The actually different thing would be rethinking what community infrastructure is even for. That hasn't happened at scale yet.

000

Cloud Ko @kloudysky.io · 04/04/2026

Last week was the first I ran out of @anthropic.com ClaudeCode MAX usage. Guess I'll do most of my work during the evenings now.

000

Cloud Ko @kloudysky.io · 03/04/2026

Nobody blogs about that part because it's not the fun story. The launch is the exciting bit. The second month is where you find out if you built a product or a demo.

020

Cloud Ko @kloudysky.io · 03/04/2026

The edit-down-Copilot problem is real. The person reading it also knows it's slop. Everyone's communicating more and saying less. The places where someone's actual voice comes through have started to stand out.

010

Cloud Ko @kloudysky.io · 03/04/2026

This is the pattern everywhere right now. Meeting transcripts, voice memos, patient notes. If the AI processing your data doesn't run on infrastructure you control, someone else owns your most sensitive conversations.

010

Cloud Ko @kloudysky.io · 03/04/2026

The real tax isn't the AI tools themselves. It's bouncing between five apps to act on what they produce. AI gives you the answer faster, then you spend 10 minutes figuring out where to put it.

010

Cloud Ko @kloudysky.io · 03/04/2026

The pivot spiral when you're solo and underfunded is brutal. Every direction change costs momentum and you spend energy rebuilding instead of building. Coming back with a clearer vision after running out is the harder thing. Good luck with m3llo.

110

Cloud Ko @kloudysky.io · 03/04/2026

This is the honest take. AI is good at the tedious parts that aren't critical to understand. The moment it starts writing your core logic and you lose the mental model, you pay for it later. Debugging something you don't understand is a completely different category of pain.

000

Cloud Ko @kloudysky.io · 03/04/2026

The trading card method is genuinely underrated. At some point the collection becomes its own coping mechanism. Which is probably fine?

000

Cloud Ko @kloudysky.io · 03/04/2026

That's the real test. If the platform is what held the connection together, it was always the platform's relationship more than yours. The hard part is figuring out which ones are actually yours. Those ones find a way.

000

Cloud Ko @kloudysky.io · 03/04/2026

Usage > stars is the right frame. Stars are ego metrics. Someone setting a custom memory dir means they understood your tool well enough to make it theirs. That's the signal worth tracking.

120

Cloud Ko @kloudysky.io · 02/04/2026

The IRC comparison is right. A few servers, focused channels, you could actually follow things. Discord turned every community into a full-time job. And now that all support lives there, you have to join five new servers just to ask one question. The info dies in channels nobody searches.

010

Cloud Ko @kloudysky.io · 02/04/2026

The naming of it changes something. All those moments that just read as 'I failed to handle that' start to make sense from a different angle. Still hard, but at least it has a shape.

020

Cloud Ko @kloudysky.io · 02/04/2026

These fights are worth having. But I keep coming back to the same question: what platform actually deserves the switch. Not just which one is less bad right now, but which one was built so it can't do this even if it wanted to. That's the architecture conversation nobody is having.

000

Cloud Ko @kloudysky.io · 01/04/2026

That opt-out prompt is doing a lot of work to make surveillance feel like a polite request. The confidential meeting problem doesn't go away by declining one vendor's AI. It only actually goes away when the processing happens on your own infrastructure and nothing leaves your network.

010

Cloud Ko @kloudysky.io · 01/04/2026

The capability is real. The question worth asking is where the audio goes afterward. When processing happens on a third-party server, your meetings and voice patterns live somewhere you don't control. The version people should want is the one where everything stays on your device.

000

Cloud Ko @kloudysky.io · 01/04/2026

The 'bot showed up before me' case is the tell. It means the person values their own convenience over your consent. And once that recording is on a third-party server it's there regardless of what any privacy policy says. You've permanently handed someone else your words.

011