A technician has to repair a machine sold fifteen years ago. The drawing exists somewhere, among network folders, binders and the computer of a colleague who has retired. On November 9 OpenAI described how a Japanese company cut that search from about an hour to a few minutes.
What happened
In its news feed OpenAI published the Kuken Industries story: the company uses the OpenAI API to search 70 years of cooling tower drawings. According to OpenAI, engineers find the drawing for a repair in 4.5 minutes, down from nearly an hour.
The most useful detail comes from the Japanese outlet MONOist, which on November 6 described the November 4 announcement made with the company Recursive. The system does two things: it takes a bill of materials (the list of components), extracts the part numbers and finds the matching pages of past drawings, even when numbering changes; then it compares customer specifications with the drawings and flags inconsistencies.
Mind the limits, which the article states:
- the 4.5 minutes come from a pilot trial, not from full-scale use;
- the 90% saving in search hours is an estimate based on fiscal 2025 data;
- the final decision stays with engineers, not the AI;
- accuracy on old drawings and handwritten notes is still being improved.
These are figures declared by the company and its partner, not an independent check.
The medieval case: Villard's sketchbook
And you will say: what does the Middle Ages have to do with this? Around 1225–1235 a man called Villard de Honnecourt filled 33 sheets of parchment with about 250 drawings: plans and details of cathedrals such as Laon and Reims, figures, machines, even a water-driven saw. Today the sketchbook is in the French National Library, shelfmark MS Fr 19093, and it is incomplete, as the Wikipedia entry notes.
Why did he make it? Historians disagree: it has been seen as a manual for architects, while other scholars read it as a memory aid for people trained orally. Almost nothing is known about Villard himself, and we are not even sure he was the one who bound the sheets.
Here is the point that interests me: an archive of drawings is precious only as long as someone knows where to look and what the marks mean. The sketchbook survived, but we have partly lost the key to reading it.
Well, your company today is in that same workshop: the drawings are there, and the key sits in the head of the person who made them.
What a small business can copy
You do not need a 70-year story. You need an archive that wastes time. Before choosing a tool, answer these questions:
| Question | Why it matters |
|---|---|
| How many times a month does someone look for an old drawing or document? | If it is rare, the saving does not justify a project. |
| Do files have consistent names and folders? | With a chaotic archive even AI gets it wrong. |
| Who checks the result? | As at Kuken, the decision stays with a competent person. |
| Where does the data go? | Drawings and customer specifications are confidential know-how. |
For more on the method, read Document management with artificial intelligence, which starts from messy archives.
What to do on Monday morning
- Measure the time lost: ask your technical office to note for two weeks how many drawing searches they do and how long each takes. Without this number you cannot evaluate anything.
- Pick a small, closed archive, for example a single product line, and let AI search it, with a technician checking every result.
- Write down who may see what: before uploading customer drawings to any service, check contracts and confidentiality.
If you want to think through how to set up the trial without losing control, see the guide to business process automation and our services for SMEs. Then book a free introductory call: half an hour to work out where to start.
The Professor of Penzai

