← Back to concept explainers

Concept Fable

Version control

Conclusion first:Version Control tracks changes to artifacts that determine system behavior. For enterprise Agents, this extends beyond code to prompts, model choices, datasets, tool schemas, policies, workflows, configurations, and evaluation suites. Version control enables auditability, reproducibility, rollback, change review, incident investigation, and controlled experimentation.

Version Control · Operations/Auditability

Read the story first. There is no rush to give a definition here; let the question grow on its own first.

Fable

A pharmacy that cannot find its old formula cannot cure new illnesses

The herbal medicine shop on Qingshi Street had a formula book. The shopkeeper changed a few ingredients every day based on patient feedback, and the apprentice would also add a few reminders in the margins. In the small shop, everyone remembered what had been changed yesterday.

After the pharmacy connected the automatic herb-picking puppet, the formula book became prompts, that set of machines, tool descriptions, the contraindication table and customer tags. There were more changes, and it was faster too.

One day the puppet suddenly began prescribing different doses for similar patients. The proprietor asked who had changed the formula, but the apprentices only remembered that someone had optimized the summary, someone had swapped the herb table, and someone else had adjusted the order of contraindications.

Naive patching means copying the current recipe and naming it "latest final version." Two days later, "latest final version 2," "temporary usable version," and "do not delete version" appear. No one knows which one is actually live.

The new medical case specialist put everything that could affect dispensing behavior into the version cabinet: formula text, herb tables, puppet models, tool permissions, test cases and release notes. Every change had an author, time, reason and diff.

They also put version numbers on the production medicine cabinets. If a patient has an abnormal reaction, they can review the formula combination used at the time; if a new version has problems, they can roll back to the previous verified version.

The version cabinet serves not only rollback but also learning. Which change reduced misdiagnosis, which change increased side effects, and which type of customer was affected can all be traced along the version history.

The pharmacy later dared to try new formulas more frequently, because each attempt no longer erased the past. Changes had a trajectory, and mistakes had a way back. The shopkeeper understood that version control is not about naming files, but about making every change in system behavior traceable, comparable, and recoverable. Later, an even more troublesome task arrived. On the surface it looked like an old problem magnified, but in fact 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 after an error. Places that used to get by on familiarity and tacit understanding now all had to be made clear. The person in charge did not urge everyone to be more careful again, but walked the site once more. He had people record every handover, every failure, and every remedy, which places were merely slow, which places would harm the results, and which places would be hard to recover once wrong.

Reveal

This story is about: version control

Version Control tracks changes to artifacts that determine system behavior. For enterprise Agents, this extends beyond code to prompts, model choices, datasets, tool schemas, policies, workflows, configurations, and evaluation suites. Version control enables auditability, reproducibility, rollback, change review, incident investigation, and controlled experimentation.

What matters is that it is not just a term, but a structural problem that repeatedly appears in real AI / Agent systems. Understanding it lets you judge when to add a model, when to change a process, and when to add governance.

Metaphor mapping

  • Recipe book: the prompts, models, tools, and policies that determine Agent behavior
  • Latest final: chaotic naming without version control
  • Version cabinet: Git, registry, or configuration version system
  • Author, time, reason and differences: commit metadata and change diff
  • Production cabinet version number: a traceable identifier for the version running in production
  • Restore the old recipe: rollback to a known good state
  • Final insight: version control makes Agent behavior reproducible, auditable, and reversible

Andy Ren Xu's judgment

The practical value of this concept is that it helps you break "AI features that look smart" into deliverable, verifiable, and governable units of work.

prompt versioningmodel versioningconfiguration managementaudit trail
京ICP备2026014272号