Industries / Finance and business services / Advertising and marketing / Graphic and web design
Layout and type
Layout is arranging words and pictures on printed pages such as posters, brochures, reports, magazines and books, and type is choosing the fonts and setting them, down to the grid of columns and where each line and page breaks.
6
Milestones achieved, from 1928 to 2013.
3
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1928
Asymmetric layout gets a rulebook
Jan Tschichold's handbook Die neue Typographie (The New Typography), published in Berlin by the German printers' education association, called for asymmetric layouts (text not centered on a middle line), type without serifs and standard paper sizes such as A4 and A5, in place of centered, decorated pages. In a lecture to Swiss graphic designers in late 1945 Tschichold turned on his own rules, calling the style useless for anything but advertising and small commercial printing, and almost wholly useless for books, in the design historian Øyvin Rannem's translation. In 1946 Penguin Books invited him to set a new standard for its paperbacks.
LessonA layout system that works for posters and advertising may not work for long reading, and its own author can be the one to say so.
Credit
Jan Tschichold
1978
Computers break lines by paragraph
In 1978 Donald Knuth of Stanford University used his new typesetting program, TeX, to set the whole 700-page revised second volume of his book The Art of Computer Programming. Its line breaker weighs every possible break in a paragraph at once to keep the spaces between words even, instead of filling one line at a time; Knuth and Michael Plass described the method in "Breaking Paragraphs into Lines" (1981). Knuth put the code in the public domain, and in the fall of 1979 the American Mathematical Society began using TeX for its own administrative publications.
Credit
Donald Knuth and Michael Plass, Stanford University
1997
One font format everywhere
Microsoft and Adobe released the OpenType 1.0 specification in 1997, building on Apple's TrueType. One OpenType font file can hold either of the two rival ways of drawing letter shapes, TrueType or Adobe's PostScript outlines, plus many thousands of characters and extras such as true small capitals and ligatures (joined letter pairs such as "fi"). Tom Rickner of the type company Monotype calls it "the now ubiquitous font standard." In September 2016 Adobe, Apple, Google and Microsoft added variable fonts, in which one file holds a continuous range of weights or widths.
Credit
Microsoft and Adobe
1999
InDesign displaces QuarkXPress
Adobe launched InDesign, a page-layout program, on 31 August 1999 against QuarkXPress, which by widely reported figures held 95 percent of the publishing market. Quark was slow to make versions that ran natively on Apple's new Mac OS X, while InDesign 2.0 did from January 2002, and by 2004 Quark's share had reportedly fallen to 25 percent, Ars Technica wrote.
LessonA near-total lead in layout software lasted only a few years once the leader stopped following the computers its users ran; the work of converting old files slowed the switch but did not stop it.
Credit
Adobe
2010
Fonts anyone may use
The SIL Open Font License lets anyone use, change and share a font; version 1.0 is dated 22 November 2005 and version 1.1 26 February 2007. Google opened its Font Directory on 19 May 2010 with open fonts that, in Google's words, "can be used for just about any purpose, including for print." By October 2023 Google Fonts had published more than 1,500 font families, almost all under that license, by its maintainers' count.
Credit
SIL International (Victor Gaultney and Nicolas Spalinger); Google (Raph Levien and David Kuettel)
2013
Layout for non-designers
Canva launched in 2013 and put page layout in a web browser, usually starting from a ready-made template. TechCrunch reported in February 2026 that Canva had more than 265 million monthly active users and over 31 million paying users in 2025, according to the company.
Credit
Canva
Watch for
Canva's design model
Canva says its Canva Design Model, launched on 30 October 2025, builds designs from separate, editable layers and objects rather than flat pictures, and in April 2026 called it "the world's first foundation model built to understand the structure, hierarchy, and complexity of real-world design." A foundation model is a large AI model trained on broad data and then adapted to many tasks. A related tool, Magic Layers, in public beta from March 2026, splits a flat single-page image into layers and restores its text as live text boxes. No outside test of the model's quality or editability was found, and no outside source backs the "first."
Achieved whensomeone outside Canva publishes a test on a public set of designs, for example how often text comes back as live, correct text and objects as separate layers.
Claimed by
Canva, October 2025 to April 2026
InDesign's AI Assistant
Adobe says the AI Assistant in InDesign can "explore, create and edit content, layouts and objects" from plain-language requests and run multi-step jobs. It has been in public beta since InDesign 21.4 in April 2026, and version 22.0 in November 2026 uses outside AI models, "including GPT 5.2 and Claude Sonnet 4.6."
Achieved whenit leaves beta and an outside reviewer measures how often it completes set layout jobs, such as setting up a 16-page document with styles and master pages, without repair by hand.
Claimed by
Adobe, April and November 2026
Print-ready posters from papers
PosterMELD, by Haojie Hu and co-authors, turns a research paper into a poster saved as an editable PowerPoint file and an image. Across 621 papers the authors report that 81.3% of requests passed their own automatic checks for geometry, readability, image quality and obvious factual errors, which they call the "Print-Ready Rate," 3.4 times the rate of an earlier system, P2P, at a mean cost of USD 0.38 per request. Design quality was scored by an AI model, not by people.
Achieved whenanother group runs the released code on new papers, gets a similar rate, and has people check that the posters are ready to print.
Claimed by
Hu and co-authors, August 2026, in a preprint
Open
01
Blind test of AI-made reports
A public test in which AI tools and human designers lay out the same 16-page report from the same brief, using the same openly licensed text and images and fonts under the SIL Open Font License. Each delivers an editable layout file and a PDF, and working designers rate the PDFs without knowing who made them.
The briefCredit · unclaimed02
A checker for typesetting faults
An open-source checker that reads a PDF exported from any layout program and flags typesetting faults, with a public test set of pages on which human typesetters have marked the faults by hand. The faults are single lines of a paragraph stranded at the top or bottom of a page (widows and orphans; sources use the two names the opposite way round, so the checker should report the position); rivers, gaps between words that line up down a paragraph; three or more hyphenated line ends in a row; loose lines with too much space between words; and headings stranded at the foot of a page.
The briefCredit · unclaimed03
Edit tests for layout files
A benchmark that gives AI layout tools the same multi-page job, then makes scripted edits to the files they deliver: add 300 words to section 2, raise the body text from 10 to 11 points through its paragraph style, change the page size from A4 to US Letter, swap a photo. It then checks whether the document reflows correctly through its linked text frames, styles and master pages without repair by hand.
The briefCredit · unclaimed
Open 01 · the brief
Blind test of AI-made reports
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test in which AI tools and human designers lay out the same 16-page report from the same brief, using the same openly licensed text and images and fonts under the SIL Open Font License. Each delivers an editable layout file and a PDF, and working designers rate the PDFs without knowing who made them.
Why it is still open
The published work found covers single pages, posters, slides or rule-based scripts, and none compares AI output with human designers' work on multi-page print documents. PosterMELD scores single posters with an AI model. OmniDocLayout generates layouts page by page and reports no human rating. Rebelo and co-authors (2023) tested an automatic book-typesetting script with people judging its output, but not against books laid out by designers. Adobe's AI Assistant is in beta with no published study.
Built with what exists
Scribus, a free layout program, with scribus-mcp, an open-source connector that "lets Claude drive Scribus" through about 125 commands, including paragraph styles, text flow, PDF export and a preflight check; a Claude model such as Claude Fable 5.2 to drive it (see the Note "Three working devices"); InDesign 22.0 with the AI Assistant beta and Canva's AI tools as other entrants; Google Fonts; text and images under CC BY or in the public domain; a simple web form for ratings.
First step
Pick an openly licensed report of about 6,000 words plus 15 openly licensed images, write a one-page brief, and see whether one AI route through scribus-mcp can produce the 16 pages.
Done looks like
At least 3 AI routes and 3 human designers each deliver the same 16-page report; at least 15 working designers rate all of them blind on a 1 to 7 scale for hierarchy, grid, type setting, image handling and overall quality; the files, raw ratings and the mean gap between AI and human scores with confidence intervals are published; every file opens in its program with live text and styles.
Open 02 · the brief
A checker for typesetting faults
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open-source checker that reads a PDF exported from any layout program and flags typesetting faults, with a public test set of pages on which human typesetters have marked the faults by hand. The faults are single lines of a paragraph stranded at the top or bottom of a page (widows and orphans; sources use the two names the opposite way round, so the checker should report the position); rivers, gaps between words that line up down a paragraph; three or more hyphenated line ends in a row; loose lines with too much space between words; and headings stranded at the foot of a page.
Why it is still open
Each existing checker works inside one tool or publishes no accuracy. lua-typo checks documents typeset with LuaLaTeX. pdf-format-analyzer, from March 2026, reads any PDF but is aimed at LaTeX output, needs a paid AI model, does not look for rivers or runs of hyphens and publishes no accuracy figures. A 2019 University of Bologna thesis built a widow and orphan checker for PDFs of four books, and its author wrote that it "reported some false positives." No public set of pages marked by typesetters was found.
Built with what exists
PyMuPDF or pdfminer, free Python libraries that read the position of each word in a PDF; rule checks for page-break and hyphen faults; river detection that tracks the gaps between words across neighboring lines; test pages typeset in Scribus, InDesign and LuaLaTeX with open-licensed fonts and public-domain text.
First step
Typeset one public-domain chapter in three programs, export the PDFs, and have one typesetter mark the first 30 pages.
Done looks like
At least 300 pages from at least 4 programs, each marked separately by at least 2 typesetters, with their rate of agreement; the checker's precision (the share of its flags that are real) and recall (the share of real faults it finds) for each fault, aiming at precision of 0.9 and recall of 0.8 for widows, orphans and hyphen runs, with rivers judged against how often the human markers agree.
Open 03 · the brief
Edit tests for layout files
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A benchmark that gives AI layout tools the same multi-page job, then makes scripted edits to the files they deliver: add 300 words to section 2, raise the body text from 10 to 11 points through its paragraph style, change the page size from A4 to US Letter, swap a photo. It then checks whether the document reflows correctly through its linked text frames, styles and master pages without repair by hand.
Why it is still open
Edit tests exist only for single designs. ReDesign's Figma Edit Replay Benchmark, with 909 Figma files and 14,796 edit instructions, tests designs rebuilt from flat images in Figma. PosterMELD reports editability for single PowerPoint posters, and Accordion relies on designers' studies of single layered designs. A GitHub search for AI generation of IDML, InDesign's exchange format, found nothing.
Built with what exists
Scribus with scribus-mcp; InDesign 22.0 with the AI Assistant beta and its IDML files; the typesetting checker above to test the results; an open-weights entrant such as Mistral Large 4 so anyone can rerun the test (see the Note "A trillion open parameters").
First step
Write 12 scripted edits for one 8-page Scribus file laid out by a person, and confirm the scripts score it correctly.
Done looks like
10 jobs with 12 scripted edits each, per tool. An edit passes when the exported PDF shows the change everywhere it should, has no overset text (text that no longer fits its frame and is hidden) and no new page-break faults. Each tool's pass rate is published with the test scripts.