Even a mathematics company needs support
A passage about mathematicians moving into field support made me think about the work that follows producing an artifact.
Proofs get releases and repairs
In Dijkstra’s invented company, proofs have releases. The first version contains bugs, and customers’ trouble reports lead to later corrections. For a story about mathematics, the product launch feels familiar.
The company in his fictional 1974 anniversary speech is called Mathematics Inc. This is satire, not the history of a real business.
A 1975 letter adds a detail that caught my attention: fifty mathematicians move from Production to Field Support to handle incoming trouble reports. People who produced new work are reassigned to care for what has already gone out.
That scene stays with me. Producing a result is visible at the front. Answering someone who encounters a question while using it sits a little further back.
It made me want to think about that later position before handing over AI-generated work.
The question after the file
Suppose AI prepares an internal analysis report. The prose is readable and the tables are filled. The file is handed over. A colleague points to one cell and asks which period its figure covers.
The question does not establish that the whole report is wrong. The work may need explanation. If the aggregation is incorrect, it also needs repair.
This is hypothetical. I am interested less in the generation screen than in the screen opened by whoever receives the question.
What do they have? Just the report file? Can they return to the source and calculation? Can they locate the part that needs attention?
A regenerate button does little to shorten the investigation if nobody knows why regeneration is needed. One question could require reading the entire report from the beginning.
Timing only the arrival of the first file would miss that later work. I return to Dijkstra’s mathematicians being moved from production into support.
Is support whatever is left over?
If I delegated this hypothetical report, I would leave somewhere for its questions to go, with a named owner who can return to the underlying material.
I would also want the responder to establish which cell in the delivered report is being discussed before producing another plausible answer. Conversations become strangely long when people have different files open.
After confirming and correcting an error, there is another job: explaining what changed. A newly generated file does not update earlier copies by itself. Someone needs to know which material to keep using.
If all of that waits for spare time, the work of producing results quickly could remain perpetually busy. Helping people use those results deserves a place too.
The empty chair at the counter
I do not want to treat the satire as a prediction that AI has fully automated mathematics. What stays with me is the odd familiarity of an organization selling results, maintaining them, and producing replacements, transposed into mathematics.
When creating more results becomes possible, I want to ask whether there is capacity for the questions that follow. I cannot prescribe an allocation between production and support here. That would require observing the actual questions and repairs.
Back at the hypothetical report, I can imagine one more sentence to attach when sending the file: where to return with a question about its contents.
The stack of finished work looks tidy. I would also like to know who gets up when the bell at the counter rings.