Anthropic has opened Claude Academy, a free curriculum built to teach people how to use AI well rather than how to use Claude's features. It's worth a look if training, not the tool, is the reason your AI rollout is stalling.
The problem it's solving
Most AI rollouts fail on adoption, not capability. You license the tool, run one onboarding session, and six months later three people use it properly while everyone else opened it twice and went back to what they knew.
Anthropic's own internal data backs this up. They found that teaching specific button-by-button behaviours goes stale quickly, because the product changes underneath the training. What holds up over time is teaching mindsets rather than steps.
Two mindsets worth stealing, whatever tool you use
The first is that today's AI is the worst AI you'll ever use. The practical implication is that skills you build now compound rather than expire, which is a reasonable answer to the very common objection that it's not worth learning something that will change again in six months.
The second is more immediately useful: verify in proportion to the stakes. A first-draft email needs a skim. A number going into a board pack needs a proper check. That single principle would prevent most of the AI mistakes we see in Australian businesses, which almost never come from someone trusting AI in general and almost always come from someone applying casual-email scrutiny to something that mattered.
It's also the sort of rule that transfers. It doesn't depend on which model you're running or what the interface looks like next quarter.
What's actually in it
Claude Academy is organised around problems rather than product tours. A legal team gets a use-case library for contract review. A marketing team gets one for campaign briefs. That framing matters, because generic training leaves each person to work out their own application, and most people don't.
It also teaches people when not to delegate to AI, which is the part most vendor training skips entirely. Anthropic frames this as protecting judgement rather than just building trust in the tool, and that's the more honest framing. A team that has never been told where the tool is a poor fit will find out the expensive way.
What stale training actually looks like
Anthropic's point about button-by-button training going stale is worth making concrete. A business that trained its team on a specific interface in March has, by August, a team confidently doing something that no longer matches the product. Worse, they've stopped looking for the better way because they believe they already learned it.
Mindset training doesn't have that decay problem. Someone who understands that a model is confident regardless of whether it's right will keep checking, whatever the interface looks like next quarter. That's the argument for spending the training budget on judgement rather than on a screenshot walkthrough that has a shelf life.
Where it fits, and where it doesn't
A free structured curriculum is genuinely useful, and it's not a substitute for role-specific onboarding against your own material. The two do different jobs:
Claude Academy builds the general mental model: what AI is good at, how to check it, when to leave it alone.
Role-specific onboarding covers your actual processes, your systems, and the decisions your team makes that no general curriculum knows about.
A governance boundary tells people what they can and can't put into a tool, which matters under the Privacy Act and matters more if you handle client or health data.
Pointing a team at the first without doing the second tends to produce enthusiasm without application. Doing the second without the first produces people who can follow a workflow but can't tell when it's producing nonsense.
The two weeks that decide it
Most of the work in a Claude rollout isn't the technical setup. It's the fortnight afterwards, where staff either build the habit or quietly revert to the old way of doing things.
What tends to work in that window is unglamorous. A small number of real tasks people do anyway, done with the tool, with someone available to answer questions. A short weekly check on what's working. A visible example or two from a colleague rather than a vendor demo, because a colleague's example is proof it works here.
A structured curriculum is a good thing to point people at during exactly that window. It gives the more motivated people somewhere to go without waiting for the next scheduled session.
What a rollout that sticks costs
For a small Australian business, a rollout built around role-specific onboarding, a written governance boundary, and a structured follow-up period typically runs A$4,000 to A$9,000 depending on team size and how many distinct roles need separate treatment.
Against a licence spend that's already committed, that's usually the difference between seats that get used and seats that get quietly cancelled at renewal. The tool was rarely the problem.
If you're rolling Claude out across a team and want the onboarding to actually stick, that's the gap worth closing deliberately. Book a time to talk it through.
