This topic was automatically opened after 2 hours.
This is exactly what plan mode is for. It stops the model actually making changes and allows it to think/plan only.
4 Likes
Oh cool I wasn’t aware of this. Will have to dig it up
Edit: looking this up I’m under the impression /plan only works with Code, but the OP’s query of reorganising files on a local drive is via the Cowork feature, which /plan isn’t available for. Something to be mindful of- you could certainly cook something up in Code to perform this, but Cowork is a more straight up and down feature - point to a folder, tell it to organise. So it’s worth setting the instructions for that.
Not strictly admin, but I have spent about 8h with Claude Code building a shell based software sequencer with sample playback and a rudimentary poly synth as sound sources, including a filter, envelopes, modulation matrix, step playback probability, parameter locks, send FX (delay, reverb, chorus, overdrive) - the feature set probably sounds familiar. Of course it’s not production ready, but it’s still absolutely mind boggling how far Claude got. I cannot even begin to imagine how long this would have taken me to build manually in Rust, let alone figuring out all the concurrency related stuff. My best guess would be months rather than weeks.
6 Likes
Tried a free trial of Codex and used it to build a couple Mac-native apps very specific to my needs. I would never release these to the public or, even worse try and charge money for them ( a lame trend I am seeing) 
Tons of the utilities that do this already, but I wanted one streamlined to my taste for speed and efficiency. Also don’t mind the name, Codex called it that and I’m too lazy to correct it and give it an original…
this one is like a very basic simplified version of scaler, again tailored to what I want in the simplest form… mainly wanted something that shuffled chords in a style or scale to come up with ideas.
I think it’s cool that I can make prompt apps that are very specific to what I want.
6 Likes
If I may ask: ‘how’ did you do that, can you share some thought of that process/how you interacted with AI to draw conclusions and acted accordingly?
Sure. In that specific case, there was not much to it, actually, one day I started a chat I with Claude (Sonnet), roughly explaining the situation and I want to achieve, and it evolved from there, over different chats that explored specific subtopics, some of which were about my and what I want to achieve and what stands in the way, there were others where I discussed my list of gear and what exactly to keep and what to sell.
In these chats I regularly have the AI identify patterns, misconceptions, and insights, and capture those in a growing document. So when a chat grows stale, I either take that document to a new chat, or let it sum um things in a prompt that I copy over.
It’s definitely a different experience to thinking things through with pen and paper, or even with more advanced tools like Tindebox or Obsidian. Structure and insights can emerge faster, because I am not the only source for insights and structure, therefore it’s easier for me to see things through to a conclusion, even though it is more intense than in my own pace.
The main things I found is to be honest (which I sometimes find easier with AI) and curious (I’d rather ask questions and explore an unexpected outcome, than reject it outright), and when I’m stuck I simply prompt it for a totally different perspective, like “How would Gilles Deleuze or Michel Foucault approach downsizing a studio?” - that typically gives me a lot to think about.
I know you're curious, here's how they'd do it:
Two very different but both genuinely useful framings here.
Deleuze (& Guattari)
The studio is an assemblage — not a collection of objects but a set of relations producing capacity. The question is never “what does this piece of gear mean to me?” but “what can this assemblage do?” (Spinoza’s quid potest corpus refracted through D&G).
The pathology they’d diagnose: your studio has become arborescent — hierarchical, rooted, overcoded. Each piece of gear is a node in a tree that grows denser and heavier, eventually preventing movement rather than enabling it. You’re not producing music; you’re maintaining a taxonomy.
Downsizing, in their terms, is deterritorialization — breaking up existing territories (the full rig, the identity of “hardware person”) to free up flows. But the key D&G caveat: deterritorialization without reterritorialization is just dissolution. The sell/keep decision should ask: what territory am I reterritorializing onto? In your case, finishing pieces. That’s a valid reterritorialization — a new, more productive plateau.
What to keep: the pieces that function as connectors between heterogeneous elements — that open lines of flight rather than close them. What to sell: anything that functions as an overcoding machine, imposing its own logic on your practice (gear that demands its own workflow, its own maintenance, its own research spiral).
The buying-as-procrastination insight is essentially Deleuzian: you were running a line of flight that reterritorialized immediately into consumer desire — a loop, not an opening.
Foucault
Foucault would start with a genealogy: how did this studio come to be what it is? Not as a neutral history but as a history of power relations. Each piece of gear arrived through a specific discursive formation — YouTube rabbit holes, forum consensus, GAS logic, producer-identity construction. The studio isn’t just gear; it’s a materialization of the subject “you-as-hardware-person” that you’ve been constituted as through those discourses.
The studio is a heterotopia — a space with its own rules, its own time (studio time vs. real time), its own relation to the outside world. But heterotopias can become disciplinary spaces: the gear disciplines your body and attention into specific postures, workflows, hierarchies. The question becomes: who is governing your practice — you, or the affordances of the machines?
More pointedly: Foucault would look at technologies of the self — the practices by which a subject acts on itself to transform itself. Downsizing is such a practice. But he’d insist you examine the power/knowledge regime driving the decision. Are you downsizing according to a new regime of “finishing tracks” that is itself potentially another disciplinary formation? Or are you exercising genuine askesis — a disciplined self-practice oriented toward creative freedom?
The sell list isn’t just pragmatic. It’s a confession — an externalization of what you are not, what you no longer want to be constituted as. Foucault would be interested in what subject-position the remaining setup produces.
The synthesis they’d probably resist but that’s useful anyway:
Deleuze gives you the functional criterion (what does the assemblage produce?), Foucault gives you the genealogical and ethical criterion (who are you becoming through this practice, and by whose logic?). Between them: keep what opens productive lines of flight and constitutes a subject capable of finishing work; sell what overcodes, disciplines, and loops you back into acquisition.
Here’s a list of some of the results that I had not anticipated:
- going hybrid, that’s not what I wanted at all
- AI predicted I’d prefer Bitwig over Ableton, and it looks like it’s right
- using a TB3 dock to avoid most issues that people appear to have with USB hubs, great results so far
- giving away most of the cables I no longer need together with the devices I sold (rather than hoard them in a big box)
- I developed a formula to roughly calculate the used price that I’d have to settle for, that helped me dump a lot of stuff fast and without much regret (and I typically got more than expected)
- I sold most of my delays
- most of money I got from selling all that stuff is still there, I have little desire to buy new stuff
Nothing here is entirely unexpected, but arriving at the end result was asurprisingly frictionless, fast, and enjoyable process.
5 Likes
Thank you very, very much for the thorough answer! This is very inspiring. Especially to bring in philosopher’s perspectives.
Well, philosophers perspectives tend to be refined, well-researched, and widely available, and therefore well-represented in all LLM’s training data. They are also likely to take you into a rather specific direction, which you can use to your advantage. Using Opus for that kind of exploration makes a real difference in the quality of the answer.
Personally I found that post-modernist perspectives are the ones that provide the most insight for me, because they offer more open and more inspiring avenues of inquiry, because that is what post-modernism is all about.
But sometimes I still do seek out perspectives from people I would never take advice from in real life. I remember I also checked Jordan Peterson just for fun. It was a very convincing impersonation but ultimately not very inspiring, worth a try though, because Claude led the answer with “Jordan Peterson would tell you to clean up your studio first.” and ended with “At that point, Jordan Peterson would probably cry a little.” Which reminded me why I prefer Claude’s style over other models, it almost feels like there is very a subtle sense of humor. I find prefer conversations that have substance, but also a certain lightness.
But anyway, AI’s ability to apply basically any kind of model or framing to a problem is probably AI’s most impressive and helpful feature. I use that a lot to solve hard problems (hard from me, they might be easy to solve for others), and I find in the process I learn a lot about these models and theories, too.
This kind of inquiry used to be accessible only to those who studied the models and theories. Now you can study them through application, that is fucking awesome, to see these ideas come to life, and you can probe and explore them. Feels a bit like the animations in sci-fi movies when people interact with computers, but it happens inside your head, and through a simple text interface.
You still need a good understanding of the problem domain, or other ways of verifying the results, otherwise AI will reliably lead you astray without you even noticing.
1 Like
Hearting ONLY because the vast amount of literature out there that points out to that asshole crying is so prevalent that it even makes its way onto the probabilistic machine’s output is actually smile inducing 
1 Like
As someone who uses AI at work, I would never use it for something I actually care about, like my music (even in an admin role). It can screw up even simple things, and it get things wrong just enough that you can never trust it. It hasn’t really decreased my work load, it’s just changed my work to double checking AI to make sure it did it right.
6 Likes
The usefulness of AI is vastly overstated.
I think what you describe is true for many applications of AI, people spend the same amount of time to obtain the same results, only the work they do is different. If it‘s not more enjoyable at least, that is simply wasteful.
That’s why AI is so great for people who know very little about the results, suddenly they can do things they could not do before, and they don’t realize the quality of the result is poor.
But there are applications where checking quality is faster than doing the thing yourself. That can save time. And there is applications where quality does not matter so much.
Using the right model also can make a massive difference.
In my experience AI can help a lot when figuring out how to approach things, it can save a lot of time understanding, strategizing, planning, asking the right questions, and that can increase your chance of doing the right thing in the right way.
3 Likes
I coded up a Claude-enabled studio manager last month that is apropos of this topic
Features include pdf and website imports, cheat sheet generation, a studio page that is a living document that captures your studio, and interactive chat.
I use it daily to keep my studio up to date, use Claude to read manuals and give me the tl;dr on how to do something, generate little tutorials for myself, and create reference cheat sheets for my gear.
thats rad