Names that already existed

If an artist meeting vibe coding for the first time feels "isn't this not really coding?", that feeling has a name that is thirty years old. Read three classic papers and vibe coding looks like an old attitude with a new tool attached.

Bricolage: Turkle & Papert, 1990 [37]

Sherry Turkle and Seymour Papert observed children and university students learning to program with Logo and Lego and distinguished two styles. Planners break a problem into parts, design modules, and then build. Bricoleurs work like a painter who steps back between brushstrokes, handling the material, negotiating with it and rearranging it.

In paraphrase: for the planner, a mistake is a misstep, while the bricoleur navigates by midcourse corrections. For the planner, a program is an instrument of premeditated control, while the bricoleur has goals but sets out to reach them in the spirit of a collaborative venture with the machine. For the planner, getting a program to work is like saying one's piece, while for the bricoleur it is closer to a conversation than a monologue.

Nine-year-old Anne's method is the vibe coding loop as it stands. She builds a simple working program and modifies it bit by bit. She starts with one black bird. She makes it fly. She gives it color. … At every stage she has a version of the finished product, not a part of it. The authors write that Anne is programming a computer but thinking like a painter. And they insist that bricolage is no stage on the way to a better form. It is one way of organizing work. The quality of the result is the same, and only the process differs.

The "conversation" that was a metaphor in 1990 became literal in 2025. One tension remains. The bricoleurs in this paper disliked black boxes made by others (they wanted to take everything apart). Generated code is the biggest black box in history. The same paper also offers a way out: bricoleurs are content to get to know other people's programs through their behavior, the way you get to know a person. That is exactly how participants in a visualization study dealt with generated code [29].

End-user programming: Ko et al., 2011 [39]

This 44-page review in ACM Computing Surveys is about people for whom the program is a means and the goal is elsewhere. It opens with the statistic that in 2012 the United States had fewer than 3 million professional programmers but 55 million people using spreadsheets and databases, and its first table lists artists who make interactive art with languages like Processing.

What this review observed twenty years ago is being rediscovered in vibe coding research.

• Requirements emerge as you build: hobbyist tinkering reconfigures technology with no fixed end point, and requirements evolve through experimentation, by seeing what is possible. → Vibe coding's "progressive requirement clarification" [7], and the observation that there are many things you would not have thought of before you saw the output [13].
• Overconfidence: 5% to 23% of spreadsheet users' judgments about values were wrong, and calling wrong values right was far more common than calling right values wrong. Immediate output boosts confidence without giving feedback on correctness. → Now the AI is a second overconfident party that also claims "done" [11][29].
• Opportunistic debugging: changing code until it seems to work leads to further errors. → Agents' side effects when fixing features [30], and beginners' "rolling the dice" [13].
• Attention investment: people allocate effort by weighing benefit, cost and risk. Thorough analysis up front can be a worse use of attention than iterative adjustment. → This is why the wiki has no separate "testing phase" and instead builds the habit of checking into the making loop.

One more important observation: the nature of the activity changes at the moment you move from a web page for yourself to a web page for others. Code pieced together without verification suddenly demands rigorous attention. That is the difference between the first-day prototype and the December exhibition piece.

Casual creators: Compton & Mateas, 2015 [38]

Kate Compton looked at tools like the Spore creature creator and Picbreeder and called them casual creators: systems that let people explore a space of possibilities quickly, confidently and pleasurably to make or discover surprising artifacts, and that give a sense of pride, ownership and creativity. The core is that they are autotelic: the goal is the pleasure of exploring, more than finishing a task. The artifact may even be thrown away once it is done.

Sentences in this paper describe the psychology of an artist sitting in front of Cursor.

• Casual users do not need full control, so they willingly trade loss of control for speed, power and surprise.
• Schön's reflective practice: see, move, see again. This cycle cannot happen disembodied in the head, and has to be enacted in conversation with the artifact.
• Design patterns: no blank canvas (the first move is the hardest, so reduce it to accepting or discarding a suggestion), mutant shopping (browse nearby alternatives instead of editing), manipulating meaningful things (let people touch "bloom", not z-scale), and the chorus (show many variations at once).

And one warning, "a thousand bowls of oatmeal": artifacts that differ to the computer but look identical to people. This 2015 name reappears in 2026 as "design homogenization" [28]. The remedy is the same too: do not accept the first plausible result, and look at several variations side by side.

By this paper's standards, LLM code generation is a rather poor casual creator. The possibility space is not narrow enough to rule out broken artifacts, the feedback is not immediate (a 30-second wait), and the default output is oatmeal. So the paper also reads as a checklist of what an artist needs to add around the tool (constraints, fast previews, side-by-side variations).

In class

Two names from 2025

Co-drifting. Krings et al. interviewed 10 practitioners and distinguished Copilot-style "co-piloting" (the human leads and the AI assists; efficiency, accuracy, reuse) from Cursor, v0 and bolt-style "co-drifting" (human and AI drift together; exploration, emergence, surprise, with code as the residue of creation). One participant said, in effect, that what they save is the flow, more than the script. The authors call vibe coding expressive end-user development for experts and non-developers alike, and propose rethinking programming as an expressive and speculative practice [9].

Interface flattening. Jin reads it through media theory. GUI, CLI and API flatten into a single prompt box, but underneath it a thicker layer forms, of remote GPUs, protocols and model providers, and control moves there instead of disappearing. The symbolic capacities of programming languages are being privatized, and expertise that used to be spread across communities and educational institutions is concentrating in proprietary systems [10]. This paper also places Processing and the Arduino IDE in this lineage. Read it as saying that if your work is to keep running after December, you need to know which layer it depends on.