Fikia / lecture
01 / 07
A Kelele Collective meetup

Make things with code.

A tour of creative coding for people who don't write code, and a working theory of where AI changes the work, and where it doesn't.

OccasionHow to Create in the Age of AI AudienceCreatives, mostly non-coders Length~20 minutes, plus a guide to keep ThesisThe syntax wall fell. Taste is the new bottleneck.
01 · The field

Creative coding is code as material, not code as solution.

The output is the point. The code is a brush, the screen is canvas and stage at once. A small map of the medium:

EX / 01

Bruno Simon

bruno-simon.com ↗

A portfolio that's also a driveable 3D world. Three.js. One person.

 EX / 02

Tyler Hobbs

tylerxhobbs.com ↗

Fidenza. His essay on flow fields is required reading.

 EX / 03

Hydra

hydra.ojack.xyz ↗

Live-coded visuals in the browser. Code as performance.

 EX / 04

Tixy

tixy.land ↗

Generative animation from a single math expression. Beauty in characters.

 EX / 05

Memo Akten

memo.tv ↗

"Learning to See": AI as medium, not tool. The bridge.

 EX / 06

OpenProcessing

openprocessing.org ↗

Where p5.js lives in public. Every sketch forkable.
02 · How LLMs actually work

An LLM is a next-word predictor. That's the whole trick.

BEAT / 01

What it does

Give it text, it predicts what word comes next. Then the next.

BEAT / 02

How it learned

Read enormous text, adjusted itself billions of times. Not retrieving: pattern-matching.

BEAT / 03

Why code works

Code is text. It has read thousands of creative coding examples. It recombines.

BEAT / 04

Two knobs

Context: what you give it. Temperature: predictable vs surprising.

Try it: pick a starting phrase, then click a word to extend the sentence.
03 · Three ways AI shows up

The model isn't one thing in your workflow. It's three.

Most creatives encounter mode one and stop there. Mode three is where the medium gets genuinely new.

I

AI writes the code

Natural language in, sketch out. Debugging. Translating between languages. The syntax wall disappears.

"draw 200 circles in p5.js, sized by perlin noise, warm palette"
II

AI as muse

You bring intent. It brings breadth and speed. Ask for ten directions, pick one, refine. The job becomes curation.

"give me five ways this flow field could be more unsettling"
III

AI as material

The artwork calls the model at runtime. A poem live. An image responding to a camera. The model is the piece.

"each second, a new line of poetry from what the webcam sees"
04 · Make one yourself

A flow field, exposed. Move the sliders.

flow_field.js · ~60 lines
05 · Prompting

Half of “AI can’t do this” really means “my prompt was bad.”

Badmake me art
Gooda p5.js sketch: 200 circles on dark, sized by perlin noise, warm autumn palette, slow drift
WhyName the medium, count, variables, palette. Specificity converts.
Badfix my code, it doesn’t work
Goodthis p5.js sketch should draw circles but draws nothing. code: [paste]. error: [paste].
WhyPaste verbatim. Models read errors well; they don’t read minds.
Badmake this more interesting
Goodfive ways to make this flow field feel unsettling: palette, motion, density, layering, timing
WhyAsk for N variations along named axes. Turns it into a real partner.
06 · Tools, compared

Which AI for which job.

Pick by what you’re doing. Fluency in one beats dabbling in three.

Claudechat + artifacts
Best for creative coding sketches. Artifacts render p5.js inline as the model writes. Default pick for this lecture’s workflow.
ChatGPTchat + canvas
Broadest capability. Strong at code, image, voice. Slightly more eager to please. Solid alternative.
Geminichat + workspace
Useful inside Google Workspace. Long context. Less battle-tested for creative coding specifically.
CursorAI-first IDE
For actual projects with multiple files. Overkill for a single sketch; transformative for real work.
Copilotautocomplete
Inline suggestions as you type. For when you already know how to code and want speed.
v0 / BoltUI generators
For web interfaces, not generative art. The right tool for the site that displays your work.
07 · The honest part

What this is bad at, and why it matters.

A talk that only sells the upside is hype. Here is what to watch for, especially this year.

↯ PITFALL / 01

The AI aesthetic

Every prompt-built piece converging on the same gradient blob, the same drifting smoke. Defaults are loud right now.

↯ PITFALL / 02

The black box you can't open

If you can't read the code, you can't fix it past the first version. Vibe-coding stalls at "okay."

↯ PITFALL / 03

The provenance question

Trained on work, including artists who weren't asked. Not settled. You need to have thought about it.

↯ PITFALL / 04

Taste does not autocomplete

The model will generate a thousand variations. Knowing which to keep, when to stop. That's the artist's job. It's the only job.

08 · Your move

Seven days. Specific moves.

The fastest way to internalize the medium is to put hands on something every day for a week.

First sketchOpen editor.p5js.org. Ask any AI for a sketch from one sentence. Run it.
Three changesOpen yesterday’s. Change three things by hand. No AI.
ReadTyler Hobbs on flow fields. Notice how much is about aesthetic, not code.
Try HydraOpen hydra.ojack.xyz. Type one line, change one number.
Read othersBrowse OpenProcessing. Fork one. Ask AI to explain it line by line.
From scratchA new sketch on a theme that matters. You decide; AI writes.
PostShare one piece. You crossed from consumer to maker. Be early.
09 · Continue

The 4,500-word version, free.

Field guide · companion to the talk almostnocode.fikia.ug/guide

Inside

  • The prompting cheat sheet: five bad/good pairs
  • Debugging with AI: the meta-skill that turns toy into tool
  • Tools, compared: six options, when to reach for each
  • Week one: the seven-day plan, expanded
  • Glossary: fourteen terms in plain language
  • Annotated resources: every link with a “when to reach for it”
End of slides. Make something this week.
←→ navigate