iiinova — where your digital self keeps evolving

A couple of new comments I didn't see before from Claude upon finalization: > Final whole-branch review is running on the most capable model over all 15 commits. Nothing else is pending until it reports; then one fix wave if needed, then wrap-up.

100

But amazingly, Claude really got good with the auto-agents - I asked it to do a spec and implementation plan, make sure every feature still works and check them. It ran for 1h 57m 57s, and the outcome now is day/night—it's blazingly fast now, instant! tbh, it was fast already before, crazy world.

100

And it just reduced the round trips for switch/screen/delete from 26 (all sequential, all blocking) to 0 blocking, with the rest running async behind the list, plus an in-memory folder cache so switching is instant. It's magical, I will test more, but so far so good.

I just did the "make it faster" prompt on my Neomd email client, as I had a slow internet connection today.

A table lists actions in an email client, showing performance metrics for before and after optimizations, with various numeric values.

Same file, no reformatting, no copying. Add the tag, run a script, and it is on the website with all backlinks. Remove friction, overshare a little, and let people correct you. The feedback is the point.

000

I do not schedule weekly reviews. Refinement is based on usage. A note I return to three times, or link ten times, is telling me it needs attention. Maybe a split, maybe a rewrite, maybe a #publish tag so it lands on my public second brain, maybe a full article.

100

Folders were the other problem. Business or personal? New folder? Is there already one? Connected notes removed all of that. Open a new note anywhere, link it to one or two others, done. Backlinks find it even if the name fails.

100

When I know I wrote about something but cannot find it, I search the content. And once found, I rename the note to what I searched for. The old name was not good enough. Next time it is instant.

100

Step two took me years to get right. The name of the note is everything. I search almost exclusively by file name, so the name has to be what I would type in a year.

100

Simon Späti 🏔️ @ssp.sh · 28/09/2026

Writing it down is step one. Not too short, or you will not know what you meant. Not too long, and never a copied article. Pause for a second, think, then one or two sentences. Often as a list, so I can indent deeper or come back up.

100

It landed at the top of Hacker News 🫣

040

Simon Späti 🏔️ @ssp.sh · 28/09/2026

Writing code by hand might be dead, but the problem isn't AI code; it's that nobody knows anything anymore.

2100

Success, by the scientist's definition: if you learned something, you did not fail. Written on vacation, inspired by Derek Sivers' How to Live.

010

The reward is long-term. Creating is hard, there is friction, it does not directly work. But it avoids the shallow happiness of the phone and Netflix, and it gives the deep flow that artists describe as the highest form of happiness.

And then release it as a gift. No return in mind, take it or leave it, no strings attached. Thinking of it as a gift removes the burden of pleasing anyone.

The outcome is a different matter. Creation is a slow process. It needs time, a slow and even boring life, to focus on details and use the craft refined over years. What matters is the quality you are proud to release.

Most of it is not conscious anyway. When we create, the subconscious drives. Let it cruise. Use the conscious mind later, for fixing errors and checking that sentences make sense. Do not start with it.

Create like a kid: effortless, exploring, seeing where it leaves you. Like play, the way kids on a campsite football field play, flawless and free. Calm like water, flowing wherever the mind goes.

120

Picasso would not have bothered to start a painting if he knew the outcome. That line stuck with me while sitting near a beach in Italy, writing without knowing where it leads. www.ssp.sh/blog/on-cre...

The Act and the Outcome of Creation

A reflection on why we create: for joy, for flow, for others, and for the self. On creation as a gift with no strings attached and success beyond outcomes.

151

Simon Späti 🏔️ @ssp.sh · 26/09/2026

4 things I learned over the years with AI: 1. If consumed by a human, it should also be written by one 2. It's hard to finish an idea that's not yours 3. If you start writing today, there's no way to know if you can write without AI 4. AI won't replace human writing (as writing is thinking) You?

171

Reposted by Simon Späti 🏔️

If you want to be a data engineer, learn these concepts. de-concepts.ssp.sh

121

I'm happy I could stay true to my principles for crafting my articles and notes, and I'm happy to do so for many more years to come. Thanks for all the support. Check out a short video showing some of the features my page supports and what it looks like. craft.ssp.sh

Data Engineering Blog & Second Brain

Top-ranked Data Engineering Blog with Genuine Insights into the Data Ecosystem with 1000s of Second Brain notes and an online book on Patterns of Data Engineering, all connected with interactive graphs.

It's all free, and the best support is if you share it with co-workers and friends. I started with data engineering, but nowadays, it has many more parts that fit my interests and life.

100

I update it many times a day. When I write my articles, when I come across something interesting, it's updated all the time (as you can see with the recent updates). Please check it out, maybe it helps you too, or someone at work.

I really think it's a resource for every data engineer out there—no matter if you start today or long ago, there's something for everyone.

Even more so, since I connected the blog more with the second brain, making them interconnected too, including my book, which made it complete. Now you can search one interface, hit Cmd+K, and search everything, all for free. brain.ssp.sh

Second Brain: Crafted, Curated, Connected, Compounded

A crafted digital vault of curated knowledge where insights and ideas connect and compound, modeled after a Zettelkasten to inspire learning and discovery.

120

But before I started my second brain, I never would have thought insights I wrote 5 years back would connect, still generate value in new content, and resurface so naturally through backlinks, links, or the interactive graph.

I have always been into long-term insight, my writing should be useful today, but also in a year or two. That's why my principles say evergreen over short-term hits, value over marketing.

Proud of what I have achieved with writing online. It's hard to consolidate writing in a form that makes sense and not being irrelevant after a week. Especially these days.

130

> Own your own website, buy a domain, and start with a static website, or if you can’t, use Substack or others, but link it to your own domain, so later you can move and keep all the links and traffic via the domain. Once you have your domain, reshare everywhere else on each social media.

Why you Should have your own Website (and Syndicate Everywhere)

Own your own website, buy a domain, and start with a static website, or if you can't, use Substack or others, but link it to your own domain, so later you can move and keep all...

011

Or, if you want to get hands-on, here is a curated list of open-source projects to choose from, from 2022 to today. The list includes more than 80 projects. oss-de-projects.ssp.sh

100

Or explore skills ranging from fundamentals (Linux commands, containerization, programming languages) to Kubernetes orchestration. The data engineering toolkit provides the building blocks of data engineering work in 2026. toolkit.ssp.sh

100

If you want to be a data engineer, learn these concepts. de-concepts.ssp.sh

121

Nothing to see here, just me traversing my little universe at ssp.sh.

011

I wrote two articles on this, the theory and a comparison of the tools. Both linked from the note.

000

Supabase branches the entire Postgres stack. Tigris forks object storage buckets like code. Turso and Guepard do instant database branching. Snowflake zero-copy clones and BigQuery snapshots are the same pattern in disguise. A contender I did not expect: SQLMesh with virtual data environments.

On the data lake: LakeFS (which acquired DVC), Nessie from Dremio, Bauplan built on Nessie, Lix for any file format, Lance for AI data. Apache Iceberg has branching and tagging built in. Inside databases: Dolt, a SQL database you can fork, clone, branch and merge.

100

That is what git gave software engineers twenty years ago. Data is catching up, and there are now two families of tools.

100

Every run executes on an isolated branch. Transformations, schema changes, ingestion all happen there. If the job fails, the branch stays open with the exact failing state preserved. Inspect it. Production untouched.

100

Engineers care about two things in production, says Ciro Greco from Bauplan: downstream consumers never see corrupted data, and nobody debugs failures under pressure. His answer is git for data. No job writes to production tables directly. www.ssp.sh/brain/git-f...