The Shakespeare BIC

A BIC Cristal can write for up to three kilometres. For the pen’s 75th anniversary, VML’s idea was to stop treating that number as an abstract claim and write a whole book with it. The book was Romeo and Juliet.

I joined the project through Bolha, where I worked as a Creative Tech Designer. Bolha was responsible for turning the idea into a working technical system. My part was the handwriting: recovering what we could from the historical references, rebuilding it as vectors and preparing it for a robot that would write the book in ink.

There is no surviving manuscript of Romeo and Juliet. The closest material we had was three pages from Sir Thomas More that are widely attributed to Shakespeare.

The scans were old, small and difficult to read. I built an AI-assisted restoration pipeline using computer vision to enlarge, clean and prepare them for analysis. It helped reveal details in the source, but it couldn’t decide what a missing character should look like. I still had to compare samples, interpret incomplete marks and work out how the letters related to one another.

From there, I built a separate AI-assisted pipeline to bridge the cleaned bitmaps into single-line vector shapes. Those outputs were working material, not finished letterforms: I rebuilt the alphabet with the team, refined the paths by hand and adapted them into a font family for the robotic writing system. Much of the job sat in that gap between historical evidence and a usable drawing instruction.

A conventional font would have caused a very literal problem. Most digital typefaces are closed outlines that get filled on screen or in print. Give those outlines to a drawing robot and it traces both edges of the letter. The result looks hollow, nothing like a pen stroke.

I built the characters as a single-line font instead. Each vector ran through the centre of the stroke and described the route of the pen directly. There was no shape to fill in later.

That changed how I drew. A curve could look right on screen and still be a poor route for the machine. Starts, endings and joins had to make sense as movements as well as letterforms.

A full book is ruthless with repeated forms. Use exactly the same lowercase “a” across 212 pages and the writing starts to look typeset, no matter how convincing that first “a” is.

I designed a variation system around multiple versions of uppercase and lowercase characters. Some glyphs had contextual forms for the beginning, middle and end of a word, so the variation could respond to where a letter appeared and how it connected to its neighbours. This kept identical forms from clustering on the page without making the text feel inconsistent.

The alphabet went through 23 iterations. Across the project, more than 22 features of the handwriting were studied, with calligraphic and forensic feedback folded back into the drawings.

At Bolha, the type work and the robotic setup were refined side by side. My responsibility was the handwriting system, but I worked closely with the people configuring and calibrating the machine. Vector paths behave differently once they become pen movement.

A point that looked harmless in Illustrator could produce an awkward join or an unnatural pause on paper. We tested the writing, watched where the motion broke down and brought those problems back into the drawings. The robot exposed weak spots that were easy to miss on screen.

The team also developed a custom holder for the BIC Cristal and calibrated the machine around the pen, paper and final page size.

The robot spent 20 days writing the first, 212-page Portuguese edition. It was presented to the Real Gabinete Português de Leitura in Rio de Janeiro, while a second copy remained with BIC.

A 226-page English edition followed and took the project to the Royal Shakespeare Company in Stratford-upon-Avon. The same BIC Cristal was used for the text and the illustrations. By the end, the book itself was the measurement of the pen’s claim.

The project received awards in design, typography, print, AI craft and creative technology. A selection:

What stayed with me was how much care it takes to build a pipeline when the expected result is not simply correct, but artistic. I built two AI-assisted pipelines: one to restore and prepare the scans, and another to help bridge bitmaps into single-line vectors. Around them were conventional processes for source comparison, drawing, path cleanup, font assembly, contextual glyph rules and testing in ink. AI accelerated parts of the work; it never resolved the design for us.

Human judgement was necessary at every handoff: deciding what in a damaged scan was credible, which proposed path preserved the character of the stroke, where a glyph belonged within a word and whether the robot’s movement still looked like handwriting on paper. The final result came from AI, software, drawing and physical testing being edited toward the same artistic standard. That is still how I approach creative technology at Bolha: build the pipeline, understand its limits and stay involved enough to protect the quality of what comes out.