Fable
After the letter-writing clerk went on duty
Stone Bridge Town originally had only three clerks. When townspeople needed contracts written, business letters answered, or old cases looked up, they all queued for them. The clerks were slow, but each person's seal, paper, cabinet, and responsibility were clear.
Later the town brought in a batch of puppets that could write letters automatically. They wrote quickly and could also search old archives. On the first day, the queue became much shorter, and the mayor thought the old process could be removed entirely.
Complexity soon appeared. The puppets sometimes cited expired clauses, sometimes mixed up similar names, and sometimes added details no one had said in order to appear complete. Worse, different puppets, different letter templates, and different filing cabinet combinations produced different results.
The mayor's patch was to paste a longer directive in front of each puppet: be accurate, be cautious, follow the rules. The directives grew longer and longer, but when errors occurred, no one knew whether the problem was the directive, the archive, the puppet version, or the review process.
After a major client received an incorrect contract, the town finally stopped to review. The old clerk said automatic letter writing is not just about buying puppets; it needs a whole workshop: who maintains the templates, how archives are retrieved, how results are tested, how costs are viewed, and who approves risk.
They established a new Correspondence Office. Each template had a version, each archive source had a marker, and each generation left a record of the machine set, prompts, retrieved fragments, tool calls, and review records. High-risk letters had to pass evaluation and human release.
The letter office also established a monthly checkup: sample quality, track latency and cost, and examine which issues frequently trigger fallbacks and which templates no longer fit the new business.
The puppets were still responsible for a great deal of writing, but they were no longer like scattered clever tools; instead, they entered a production line that could be deployed, monitored, evaluated, governed, and rolled back. The mayor later understood that the large-language machine could write answers, but could not automatically bring an operable system to the enterprise; the truly difficult part was delivering answers stably, controllably, and explainably in the business over the long term. Later, an even more troublesome task arrived. On the surface it looked like just an amplified version of an old problem, but in practice it forced out all the gaps hidden in the process: who takes over first, who can make changes, and how to return to the previous step when something goes wrong. Places that used to get by on acquaintances' tacit understanding now all had to be made clear.