Most product changes do not invalidate any training
The product is constantly changing, and training becomes obsolete without anyone noticing. Organizations either review all their materials every quarter or discover the gap when a customer tries to follow an instruction that no longer works.
Most changes involve nothing at all. Renaming an internal field, improving performance, or introducing a beta flag does not make a module incorrect. What matters is a change to something that a module tells a person to do, such as moving a control, removing an option, or altering the sequence.
To make that decision, you compare the change against the module's instructions rather than its subject. Modules covering the same feature can be affected in different ways. The module that is usually affected is the one with many unexpected screenshots.
The agent creates a list of modules to revise, with the changes and relevant passages. Training owners then decide which parts should be rewritten. A person publishes the revisions.
How the agent is built in Agent Studio
Prompt skills state the release sources: changelogs, repositories, and release notes. The agent draws conclusions only from these sources, not from any inference about what a change probably achieved.
The second prompt skill keeps the module inventory as instructions rather than titles, because a title does not indicate whether a moved button breaks the module.
The system prompt requires a reference to the change and a quote of the instruction it contradicts. If a module is flagged without a quoted passage, it is listed as uncertain.
The agent has to state when a change has no effect, because a quiet result is a finding, and silence is taken to mean failure.
A Start Scheduled flow reads GitHub releases. An HTTP Request node retrieves changelogs. Both pass the module inventory from Notion to the Agent Chat node.
Each affected module receives one revision task from Monday or Linear, with the change and the relevant passage, so the work is already specified on arrival.
The agent does not edit any modules. A Human in the Loop node holds every revision before it reaches the learners.
What this agent is built from
- Prompt skills: declared changelogs, repositories, and release-note sources.
- Prompt skills: module inventory held as instructions, not as titles.
- System prompt: each flag cites the change and quotes the contradicted instruction.
- System prompt: a period with nothing affected is reported as such.
- GitHub (Integration) with HTTP Request (Util): release and changelog collection.
- Linear (Integration): one revision task per affected module, already specified. Monday fits the same slot.
- When an ActionFlow is enough: If flagging modules whose feature names appear in a release is the work, build the workflow. That flag is a usable first filter.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.