Ramp time for new hires is dominated by an unbudgeted line item: learning the software stack, the eight tools and forty workflows a role actually runs on. It's taught today by interruption, new hires borrowing seniors' time question by question. An AI trainer replaces that with a role-based agenda: spaced sessions, supervised practice, and a measurable finish line per tool.
Where ramp time actually goes
Strip out the org chart tours and the benefits paperwork, and most of a new hire's unproductive weeks are tool acquisition: which system does what, how this team's configuration differs from the last job's, where the process lives, what the folklore rules are. The learning channel is social: watch a colleague, ask when stuck, absorb by proximity. It works, eventually, at two costs nobody meters: the senior team's attention (the best people field the most questions) and consistency (each hire learns a different subset, with the local shortcuts baked in from day one). Ramp isn't slow because people learn slowly. It's slow because the teaching is rationed and randomized.
The ramp agenda: a curriculum per role
Treat the stack like a curriculum. For each role, list the tools and, per tool, the workflows that role must perform unassisted, sourced from the people who do it best. Sequence by dependency and frequency: the daily workflows in week one, the monthly ones before their first month-end, the rare-but-critical ones in practice mode before they're ever needed live. An AI training agent runs this agenda in short spaced sessions on the hire's own screen: demonstrate, practice on realistic cases, verify, resume where the last session ended. The new hire's question stream doesn't vanish, it redirects: "how do I" goes to the always-available trainer first, and seniors get the questions that deserve a senior.
Practice before production
The highest-anxiety moments of ramp are first performances: first ticket, first quote, first payroll run. A practice mode converts each into a rehearsed performance: the hire has already run the workflow, including its common exceptions, under supervision, on realistic data. First-week errors stop being the tuition fee for learning the stack, and managers stop choosing between throwing hires in early (errors) and holding them back (waste). The finish line per tool is explicit: core workflows performed unassisted, witnessed, recorded.
Measuring ramp like you mean it
Replace "feels ramped, roughly a quarter" with instrumented milestones: time to first unassisted execution per core workflow, per hire, per role. The metric surfaces where ramp actually sticks (invariably two or three specific workflows), turns manager intuition into a curriculum fix, and gives every next class a baseline to beat. Watch the senior-team side too: interruptions displaced is real capacity returned. (Disclosure: we build Skippr; Skippr AI training runs exactly these ramp agendas, sessions plus practice mode, measured on users trained, and it starts self-service, so piloting on your next hiring class is a small experiment, not a program.)
Questions buyers actually ask
How much of ramp time is really software training?
Audit a week of a recent hire's questions: the majority typically concern tools and process mechanics, not domain judgment. That share is addressable with structured in-tool training.
Do buddies and mentors go away?
No, they get promoted: freed from "where's the export button," mentor time concentrates on judgment, context, and culture, the things only humans teach well.
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.