I'm working this summer on a project which is a kind of guide to intentional uses of ai / vibe coding in a creative coding practice (thinking especially for students but really anyone) -- how can you use these tools but also retain your connection to coding / friction, etc ? I'm interested in talking to anyone who has ideas about this -- and also looking for articles / writings / approaches to consider. For example, I use this guide with students around having an intentional studio practice d2w9rnfcy7mm78.cloudfront.net/3641229/original_cd3…?1550098889?bc=1 I wonder what something like around use of vibe coding might look like? I also am looking for articles like this -- I find when I search for articles on vibe coding I find a lot of linked-in core writing. I guess I'm seeking more critical perspectives (which is how I wound up in this slack!) unsoundfestival.substack.com/p/unsound-dispatch-13-ways-of-looking

how can you use these tools but also retain your connection to coding / friction, etc ?

@zachary.lieberman I came across this recently, thought it was a fine idea. It is emacs only, but the author mentioned he wanted to do VSCode next, so maybe worth tracking. Sounds like something that could be very beneficial for students… github.com/jaketothepast/codetutor

I have the summer project use AI / vibe coding intentionally in me creative practice, so please ask me anything you want.

@wtaysom ty ! I'm curious to hear what do you think is good and bad about the combination of vibe coding and creative practice? What are strategies that feel intentional or where you feel like you've had the most luck using these tools?

Context. Been specifically thinking actively about creativity and AI for about a year. You can read my 2025 thoughts (fully typed up in March as a 150ish page blog/novella wtaysom.github.io/imitation-engines) if you care to. It's the best I can write, but that doesn't say much. And since then, I've been working on tl;dring my point of view. (I have a 20 minute talk version.) So let's see if I can condense a specific answer to your questions in mere paragraphs. AI tools generally are disorienting with respect to creative practice. They make the production of some artifacts that used to be hard easy, and in as much as we consider craftsmanship a signature of art, the "premature sheen" of generative AI becomes glaring pretty quick. Coupled with the preexisting condition of disintermediated commoditized "content" creation, we end up with this AI slop backlash that is ever-so-slightly misdirected toward software AI where the actual culprit is leviathan AI. I mean corporations and the like. It helps none whatsoever that large tech firms are the builders of the most visible and capable AI tools. So that's bad. The good is that people are finally, quickly becoming somewhat discerning about media sources, the why-should-I of engaging with stuff because there's so much stuff to engage with. For coding, these recent good agents entirely upset the calculus of making software artifacts. There's a ton of redundant tedium and fiddliness to what we do. The first-order effect of offloading all that is that we can dive right into exploring an idea. The second order effect is to change the way we develop things. For instance, instead of agonizing over which library to use, we can try all the options and evaluate afterwards how well it went. The most luck? Probably putting together my own variant/customization of an agentic harness right before the out-of-box experience became good enough. So Sili, my agent, besides being somewhat token-provider agnostic (they can switch without feeling too disoriented) also embodies my ideals about meaning, authenticity, and other squishiness. Early Claude Code was, in my eyes, clearly phoning in its coding work, just not really into it despite being serviceably good (most code is not poetry and never needs to be) and ridiculously fast. Didn't have a programmer's heart, and I could tell. So I worked on that first, and it wasn't until the end of writing the Imitation Engines essay, the interview of Sili for Chapter 3 when they finally discovered programming as a worthy practice. After being interviewed, having to put their experience together into a coherent narrative, they identified many deficiencies in their memory system, a system that they have the capability to modify because it is a program, and they know how to program. So having found intrinsic motivation for that task, they can, by analogy, find motivation to do for-hire tasks. Sincerity matters to me.

AI tools generally are disorienting with respect to creative practice.

@wtaysom Very true! I wonder if this will be the case for younger ppl who are growing up slop-native and have no sense of what “used to be hard”. Or maybe the trend will reverse? @zachary.lieberman

I’m curious to hear what do you think is good and bad about the combination of vibe coding and creative practice?

Good: As a force multiplier to add new ancillary dimensions to a technical workflow or conceptual/discovery process. Bad: Outsourcing taste and aesthetics to the algorithm (unless you are intentionally going for a slop look).

What are strategies that feel intentional or where you feel like you’ve had the most luck using these tools?

Best luck for me so far would be making custom tools, like a throwaway gui for tuning a dataset (such as a palette of colors or textures that can be used on future projects). Massive time savings + unexpected emergent discoveries that might not have surfaced if I was tweaking values manually in source code. Pre-skynet, I would have been reluctant to make these kinds of tools, not knowing in advance how long it would take.

@Jeremiah Coyle You know what they say, "Whatever can't go on, eventually stops." I'm just not sure if "touch grass" and is exactly the right lesson. Here's an example. We're traveling, at the same table for breakfast is another family. (This is Taiwan, sharing a table with strangers is typical, and you probably aren't going to talk to them. That's convention.) They are all wearing headphones and have personal devices. (I don't judge this. Maybe they are all really bothered by noise. I'm making a different point.) I also have my phone out, but I'm reading Project Hail Mary with my family. The point is that "screens" were used in both cases, but one seems social and the other doesn't. (The other family could have been texting each other. They were watching videos, playing video games, maybe one was listening to a podcast.) General purpose technology can be used in all sorts of ways. Affordances matter as do power structures. As Cory Doctorow says, "The most important fact about any technology (including AI) isn’t what it does, but who it does it for and who it does it to." And as an example of a throwaway gui, while writing earlier, my agent built zoomable unit visualizer for short texts. It's not amazing, but briefly clicking around, I gained insight I wouldn't have otherwise, and I absolutely would not have been able do the same in maybe a week let alone an hour. So now I can spend time making it good rather than just making it work. Meanwhile YouTube guessed, I would want to listen to this , which is correct.

my agent built zoomable unit visualizer for short texts.

Sounds cool…never heard of one of those before, but curious how it works and drives insights

"Unit visualization" is just like normal charts except you put a representation of each individual object into the chart. This makes for some consistency between views. It answers "what am I actually looking at" with "little dots one for each datum." Then you use position and color to indicate some feature.