Until this redesign, every Design Outcomes hero was an illustration, mostly risograph prints and Dribbble work, each one cleared with its artist by email before the issue went out. I love that work, and I also collect art, from old paintings to modern prints, so I wanted the publication to carry that side of me: one painting per article, chosen because it says what the article says.

The idea came from Goya. His painting of Saturn devouring his son is the most literal picture of feature cannibalization I can think of, and everyone knows it on sight. The work was finding that kind of match every week, and writing a line that lands on both the painting and the article.

The system is a scheduled task in Claude Code plus a set of rules in the production file it reads before it writes anything. It reads my week of call transcripts, drafts one to four articles in my voice, finds a painting for each, writes the headlines, builds the issue on its own branch and leaves it running on localhost.

Then it stops, because the parts that need taste are mine. I read the issue where readers will read it, send edits and the drop caps I letter myself, and nothing ships until I say go.

The first instinct for any metaphor is the most famous painting that fits it. My articles are mostly about people doing something, so the metaphors reach for paintings of people doing something, like Thomas doubting or a maid delivering a letter, and that steers straight to Renaissance and Baroque story painting. Issue 24 landed in three era bands out of seven.

I called it out, and the fix went into the production file: seven era bands, at most one piece per band, and at least one piece from outside European oil painting whenever one fits. On the re-pick, Issue 25 went from four bands to six, with a Muybridge photograph and a Hokusai woodblock taking the slots a Hobbema and a Bruegel copy had held.

Museum records come first: The Met, the Art Institute of Chicago, the National Gallery of Art, the Smithsonian and the Rijksmuseum. For anything after about 1900, the museum’s own public-domain flag is the only thing that clears a work, and it changes piece by piece: the Art Institute marks Kandinsky’s 1913 Improvisation No. 30 public domain and Matisse’s Bathers by a River not. Wikimedia Commons supplies the file only when a museum’s server won’t, and only files marked public domain or CC0.

Tantara is my iPhone app that gives a photo a sitcom title-card introduction, written by a three-pass writers’ room: a literal read, a context read, then five titles in five different devices. Each painting goes through that same skill, with the article’s thesis passed in as context.

The room judges the painting, so its own pick isn’t enough. Every candidate also has to read as a caption for the art with no article attached, and as the article’s thesis with no art attached. A line that passes only one of those doesn’t ship.

The reveal reads its timing from the same constants as the video reels in Tantara, so the two can’t drift apart. The reel holds its black for almost six seconds, which suits a video someone chose to watch. A reader is waiting on a painting, so I asked for half: the black starts to lift at 1.92 seconds and the painting is clear at 2.82.

The painting always shows at its own proportions, centered, with a blurred copy of itself filling the rest of the screen, and the headline sits in a dark band so it reads on any canvas. The card waits for the painting to load before it lifts, so it never fades onto an empty frame. Raphael’s School of Athens taught one more rule: it’s a fresco shaped like an arch, scanned onto a white rectangle, and the corners showed up as white wedges. A small script now clears the surround of any work that isn’t a rectangle before it ships.

My mockup put the headline and the “why it matters” on the left, and the week’s paintings on the right, hung salon style. The piece you’re reading sits in a black frame that turns white, clockwise, over six seconds, and then the next piece takes over. Point at any painting and the copy follows it; click and you’re in the article.

I hang walls so every filler lines up with the art beside it, and the first two hangs didn’t. The third borrowed the justified-row mosaic from my Illustrations & CommD page, which lined everything up and filled the box completely. A good salon wall has some whimsy to its layout, so for the fourth I sent a reference wall: the hang clusters around an anchor and leaves its outline ragged, and every edge still lines up with a neighbor. I tried white blocks at the ends too, and the wall looked better with them gone.

The strip under every article used to be a row of thumbnails. I sent a sketch with Goya’s Saturn in the middle and three paintings worked around it, and that became the rule: the piece you just read hangs in the center, larger, and the rest of the issue sits in columns of one or two, each centered on it, with paired pieces sharing the edge that faces the centerpiece.

I title photos for Instagram with Tantara almost every day, so every title I publish is the clearest signal I can give about the wording I like. The app now keeps each published title with the four I passed over, and every new request carries them, so the room leans toward the length and temperature I keep choosing. Nothing is stored on a server, which kept the app’s privacy promise intact.

I didn’t write any code. Claude did the searching, the license checks, the skill runs, the layout math and the verification on every build. My part was setting the bar in the brief and reacting to what I saw on localhost, and each reaction became a rule the system keeps, so the next issue starts where my last note left off.