17 name cards in one evening: the AI-assisted print pipeline

My mother turned 83 this month, and the family lunch needed 17 name cards on the table. Three years ago I had designed a card set for her 80th — and the source file was gone. What followed was one evening of work that would have taken me a full weekend before, and most of it was delegation, not design.

Reconstructing the lost original

The first problem wasn’t making new cards. It was finding the old ones. The agent searched the file archive, found references to the 80th birthday set, located a printed sample, and measured it: spacing, cap height, letterforms. That led to a font identification (Avenir Next, medium weight, tracked wide) and — more importantly — a measured tracking value calibrated against the original: not “about 250 thousandths” as a guess, but tested against renderings until the match was right.

That’s the difference between AI as a design tool and AI as a production assistant. I didn’t want a new design. I wanted the old one back, verifiably.

The pipeline

1. Guest list from history. The agent pulled the list from the planning notes: 15 adults, 2 children, one name appearing twice (two Hannes) — resolved with initials, the same convention as last time.
2. Per-guest data. The two youngest guests had pre-ordered children’s menus; one guest was vegetarian. The cards encode that — each guest’s menu choice sat in the aperitif line where a wine pairing would normally go. (We decided early: no wine section on the cards. People pick their wine at the table.)
3. Print-ready layout. A4 landscape, three cards per sheet, 9.5 × 21 cm cards, cut marks, 100% scale, six sheets. Delivered as PDFs plus a vector skeleton for the layout app, in case I wanted to restyle anything by hand.
4. The print itself. I printed them at home. Total hands-on time for me: reviewing the proofs and feeding the printer.

Where AI actually helped — and where it didn’t

The heavy lifting was document archaeology and layout math: finding the old template, measuring it, generating 17 individually correct cards from structured data, and getting every sheet right the first time. That’s maybe 80% of the work, and it’s exactly the kind of tedious precision work AI is good at.

What it didn’t do: decide the design. Whether the cards should carry a wine section, how the vegetarian’s card should read, whether “Hanne” needs a distinguishing initial — those are host decisions, and they’re also the quick ones. The agent asked, I answered in a sentence each, and the pipeline absorbed it.

The general lesson

If you have a previous instance of almost any family document — invitations, menus, cards, programs — you’re one evening away from a repeatable pipeline. The expensive part of print work is never the layout software. It’s the reconstruction of what you did last time and the data hygiene around the guest list. Both are exactly what an AI agent is good at, and neither requires trusting it with taste.

And the practical result: 17 guests, 17 correct cards, zero misprints, and the file — this time — archived with a checksum so the next birthday starts from a working template instead of a lost one.