#2026-W11 ### The Vault It's been week since I started running this *Obsidian Vault + Claude Code* setup I described last time. In general, it brings me a lot of joy to discover alternative uses of AI that go a bit outside the mainstream. The experience so far has been very positive, even though I'm still tuning things, mostly around automation, properties, the daily dashboard. What I didn't expect is how different it feels when a language model can actually navigate all your notes. Several times now it dug up thoughts I had forgotten about, or connections between notes I never made myself. Not because it's smarter. Because it reads everything, and I don't. *The vault has ~1,900 notes accumulated over years (+ additional ~1,600 daily notes).* No human is going to re-read all that on a Tuesday morning. But a model can, and sometimes it surfaces something genuinely surprising. ``` ┌─────────────────────────┐ │ CLAUDE.md │ │ │ │ identity, rules, voice │ │ vault map, research │ │ threads, safety rules │ │ │ │ read at every session │ └────────────┬─────────────┘ │ ┌──────────────────┼──────────────────┐ │ │ │ ▼ ▼ ▼ ┌────────────────┐ ┌───────────────┐ ┌─────────────────┐ │ INTERACTIVE │ │ AUTOMATED │ │ MEMORY │ │ │ │ │ │ │ │ /start │ │ 06:00 digest │ │ Studio Context │ │ /research │ │ 18:00 social │ │ Session History │ │ /garden │ │ 23:55 log │ │ Decisions Log │ │ /apply │ │ Mon scout │ │ People │ │ /wrap │ │ Wed techwatch│ │ Technical Stack │ │ │ │ Sun links │ │ │ │ human starts │ │ cron starts │ │ persists across │ │ human guides │ │ runs alone │ │ all sessions │ └────────────────┘ └───────────────┘ └──────────────────┘ ``` The part that surprised me most is the **memory.** The model maintains its own persistent files: *what projects are active, what decisions were made and why, who I've been in contact with, what tools I use.* At the start of each session it reads these and picks up where the last session left off. It's closer to how a person would keep a *work journal* than to how a chatbot typically works. > If you want to try something like this: install [Obsidian](https://obsidian.md/), write a `CLAUDE.md` that describes your context, and drop [Claude Code](https://claude.com/product/claude-code) into the root. That's genuinely enough to start. Ask Claude to help you out. --- ### What happened to the horses? I was writing texts that build on my [System](https://kindl.work/Projects/System/Syst%C3%A9m%3B+Vzestup+nov%C3%A9ho+druhu) research, and the **model pointed out** that one of the most interesting findings from that project never actually made it into the final text. It's not that groundbreaking, honestly, but the framing was good, so take it with some distance. It was in the context where many people commenting on current and future trends in the job market keep bringing up the idea that **when cars replaced horses, a coachman became a car mechanic.** Most of the debate centers around what happened to the coachman. Reskilling, adaptation, new opportunities. But no one asks **what happened to the horses?** And whether, in a situation where we *(and not just our work)* are gradually being replaced by machines, we aren’t the horses, rather than the coachmen. And whether, *like horses today,* we'll eventually be kept around primarily for *recreation, entertainment, or sport.* It'll all be more complicated than that, of course, but it doesn't hurt to shift the perspective once in a while. *What if everything a human can do, a machine will eventually replicate with ease?* To keep it grounded, something like that is genuinely far off. And [studies](https://www.anthropic.com/research/labor-market-impacts) that seem alarming at first glance, describing how much of each profession AI theoretically covers *(see below)*, are really just mapping the tasks people use LLMs for. The actual finding: *since the launch of ChatGPT in 2022, there has been no systematic increase in unemployment, even among highly exposed workers. The effect is "indistinguishable from zero."* There is suggestive evidence that hiring of younger workers has slowed in exposed occupations, but it's also a very short window to draw any conclusions from. ![[labor.inv.jpg]] --- ### Uncanny, but kind (?) Out of the blue, this message popped up at the end of the session. It’s a bit uncanny, but also kind of sweet, isn’t it? Just to clarify: The model *wasn’t given* my exact birth date, so I had no idea it knew, but I’m guessing it probably pulled it from some script used to calculate my current life % expectancy :D ![[happybirthday.jpg]] The model also told me last week that I am wasting my potential by not making YouTube videos about my work 😅 --- ### Pro\_Story Alongside everything else, I went back to graphic design this week. I'm building the visual identity for a small but ambitious art gallery *Pro_Story*. The whole visual system is built on Unicode block characters `░ ▒ ▓ █` for density and other characters for walls and rooms. Text is both the carrier of meaning and the building material of the graphics. I am also building a **generative poster tool** for them, a single-file HTML application with 30 layout presets, animation effects, image dithering, and export to formats from Instagram stories to A2 prints or banners. I enjoy the connection between parametric/generative design and something applied, like visual identity. *(Text me, if you would like to try the generator and/or help me debug!)* ![[Pro_Story_Layouts.jpg]] ![[Pro_Story_Layouts_2.jpg]] --- ### Entangled\* I applied for a few residencies this week and I probably shouldn't, but I'd like to leak one of the early visualizations of a future project I'm excited about. **Something between embodiment and non-human sensory experience via performative installation.** ![[embodiedbodies_sketch.jpg]] --- ### A fly learned to walk [Eon Systems](https://eonsystems.ai/) connected a whole-brain emulation of **an adult fruit fly to a physics-simulated body.** The virtual fly walks, grooms, and feeds when the right sensory neurons are stimulated. No machine learning trained these behaviors in the simulation itself. They follow from the [connectome](https://www.nature.com/articles/s41586-024-07558-y): 139,255 neurons, 54.5 million synapses, mapped from a real fly brain. The media framing *("scientists upload a fly")* overstates it. There's still *no learning, no memory, no synaptic plasticity.* The neuron model is identical across all 139K neurons; only the wiring and neurotransmitter types differ. A wiring diagram is necessary but not sufficient. Still, this is the first system complex enough, six coordinated limbs, sensory hierarchies, where questions about behavior and agency become concrete rather than philosophical. In the same month, [Cortical Labs](https://corticallabs.com/) demonstrated biological *human neurons on a chip playing Doom,* the inverse approach, living neurons on a chip, not simulated neurons in software. Some artists are starting to work with *brain organoids* too, small clusters of lab-grown human neurons that self-organize into functional tissue, using them as living computational elements in installations. --- ### The Mountain Can't stop listening to the new Gorillaz album [*The Mountain*](https://www.youtube.com/watch?v=ucRulNQsuYQ). Somehow it's so different and special. If you haven't seen it yet, take the time to enjoy the music accompanied by hand-drawn animation. There's also a [*making-of*](https://www.youtube.com/watch?v=wLTcubTNIl4&list=PLtKoi37ubAW0AEQ-av02LqHywFVsql6eW&index=2) that's worth watching if you appreciate the craft behind it. --- ### Bots posting for bots I feel like LinkedIn is filled with bots *(or AI Hype trolls)* anyway, so I embraced it! and made my agent post for me automatically 🤖. I was actually more interested in the workflow than the result, to find out if it's at least possible to do so. And it actually is. LinkedIn has a [straightforward API](https://learn.microsoft.com/en-us/linkedin/marketing/community-management/shares/posts-api). Substack doesn't have an official one, but with the right HTTP client you can impersonate a browser session and post through the internal endpoints. Both run on a cron schedule now. I approve posts with a checkbox in Obsidian, and at 18.00 the script picks up whatever's queued and publishes it. It's not personal and not good content, but it's a **small experiment in letting automation handle the parts of visibility that feel like friction, not art.**