Every training program has a second, unpaid job: maintenance. The product ships, the screenshots rot, the course drifts out of truth, and some percentage of your curriculum is quietly teaching last quarter's interface. The evergreen classroom is the alternative: training that derives from the live product, so when the product changes, the training already has.
Content rot is a schedule, not an accident
Training content doesn't decay randomly, it decays at exactly the speed your product improves. Every release note is a small obituary for some screenshot, some video segment, some carefully authored module. Teams respond with refresh cycles, content audits, and "last reviewed" spreadsheets, which is to say: with a second content operation whose entire output is keeping the first one from lying. The faster you ship, the bigger the janitorial staff. It's nobody's favorite budget line, and it's structural to any training model built on artifacts, because artifacts are snapshots, and the product is a stream.
What "evergreen" actually requires
An evergreen classroom can't be a better-maintained library. It has to invert the relationship: instead of humans re-describing the product to the content, the training layer perceives the product directly. Three properties follow. Self-updating knowledge: the trainer rescans the product and docs as they change, so the curriculum's source of truth is the current interface, not a render from March. Live delivery: lessons happen on the learner's own screen, in the product's present tense, there is no screenshot to rot because there is no screenshot. Always-on availability: the classroom isn't an event you schedule but a colleague you summon, mid-task, mid-question, mid-release. That combination is only possible when the trainer is an agent with eyes on the product rather than a playlist in front of it. (Disclosure: it's the design behind Skippr's Skippr AI training, brain that rescans, sessions on live screens.)
What the maintenance budget becomes
Not zero, someone still curates what's worth teaching, sets the agendas, and reviews what the agent learned. But the work changes species: from re-rendering artifacts to editing judgment. The release-day scramble ("who's updating the training before Tuesday's launch?") becomes a non-event, because launch-day training is just the agent teaching the product as it now is. Your enablement calendar stops orbiting your release calendar. That, more than any single feature, is what evergreen means: training that is never between versions.
Watch it train someone
Completion is not competence, and the difference shows up in the product rather than the dashboard. Fifteen minutes is enough to tell them apart.