On 21 September 2026 OpenAI expanded OpenAI Academy with role-based learning paths. There is now a track for knowledge workers, a track for developers building with Codex and the API, an AI Leadership course for executives, and courses for educators and students. It is a tidy, well-organised library, and for a lot of staff it will be a perfectly decent first look at prompting and reviewing AI output.
The question an Australian operations or engineering lead should ask is narrower than "is this good content?" It is: what does my team need to learn before AI is doing real work in our business, and can a self-serve library teach it? For most mid-market teams adopting Claude Code, the honest answer is that a library covers the first week and very little of what comes after.
Is OpenAI Academy enough to train an Australian team on Claude Code?
No. OpenAI Academy teaches general AI skills and OpenAI's own tools, so it cannot cover Claude Code, your repositories or your workflows. It is a reasonable free primer on prompting and checking output. Teams adopting Claude Code still need hands-on training run against their own codebase, their review process and the specific tasks costing them time.
That is not a knock on the content. A generic course has to be generic. It cannot know that your billing service has no tests, that your team reviews pull requests in a particular way, or that the real time sink this quarter is migrating forty reports off a legacy system. Those details decide whether an AI coding tool earns its keep, and they only show up when training happens inside the work.
Course library or working rollout: two different purchases
It helps to treat these as separate things you are buying, because they solve different problems and are measured differently.
A course library
Hands-on Claude Code training
A rollout
Most teams that report "we did the AI training and nobody uses it" bought the first and expected the third. We covered the same pattern with Anthropic's own curriculum in our piece on Claude Academy: good material, but it does not carry a team through the fortnight after launch.
| Option | What it teaches | Measured by | Typical gap |
|---|---|---|---|
| Self-serve library (OpenAI Academy) | General prompting, OpenAI tools | Course completions | Nothing tied to your code or data |
| Generic Claude workshop | Claude basics, safe use | Attendance, feedback | Little carry-over into daily work |
| Claude Code training on your repo | Context files, review, task scoping | Pull requests merged with Claude | Needs a senior engineer's time |
| Claude rollout with training built in | Named workflows end to end | Hours saved per workflow | Higher upfront cost |
What Claude Code training covers that a library cannot
The skills that decide whether Claude Code sticks are specific to the codebase it runs in. In practice, the sessions that change behaviour spend most of their time on four things.
Project context
Claude Code works far better when the repository tells it how the project is built, tested and reviewed. Writing and pruning that context is a skill. A generic course cannot teach it because the right content depends on your stack, your conventions and the mistakes your team keeps making.
Scoping a task an agent can finish
"Refactor the payments module" fails. "Move these three functions behind an interface, keep the tests green, stop before touching the database layer" usually works. Learning to cut work to that size is the single biggest gain we see, and it is learned by doing it on real tickets.
Reviewing agent output
Developers need a review habit for changes they did not type. That means reading diffs for intent, running the tests locally and knowing which files should never be touched without a human. Your existing code review rules are the starting point, not a course template.
Permissions and guardrails
Which commands Claude Code may run unattended, which secrets it must never see and who approves changes to shared infrastructure are governance decisions. For teams in APRA-regulated or ASIC-supervised businesses, those choices also need to line up with existing change-management and Privacy Act obligations.
What it costs to get this wrong
The cost of training that does not stick is mostly salary time, not course fees. An illustrative example for a 12-person development team in Melbourne:
Each developer spends 6 hours on a self-serve track: 72 hours of engineering time, roughly $9,000 at a loaded rate of $125 an hour.
Six weeks later, three developers use an AI coding tool daily and nine have drifted back to old habits.
The licences keep renewing for twelve seats while value comes from three.
Compare that with two days of hands-on Claude Code training on the team's own repository, followed by a short check-in at week six. The fee is higher, often in the $15,000 to $25,000 range for a team that size, but the measure is merged work rather than badges. These figures are illustrative; your numbers depend on team size, rates and how much of the codebase is in good shape. Our ROI calculator is a quick way to test them against your own payroll.
Where OpenAI Academy still fits
None of this means ignoring free material. A library is a sensible way to give non-technical staff a shared baseline before a rollout, and the AI Leadership course may help executives frame the decision. Use it as pre-reading. Do not count it as the training plan for an engineering team adopting Claude Code.
If your organisation is weighing whether to train existing staff or hire, our upskilling versus hiring breakdown runs the Australian numbers. For the formats that change behaviour once training starts, see our guide to AI training formats.
A practical sequence for the next quarter
Week 1: pick three repetitive engineering tasks that cost real hours each sprint.
Week 2: set up Claude Code against the repository, with project context and permission rules agreed.
Weeks 2 to 3: hands-on sessions on those three tasks, with pairs reviewing each other's agent output.
Week 6: count merged changes and hours saved per task, then decide what to expand.
If your team has done generic AI training and still is not using it day to day, that is usually a rollout problem rather than a knowledge problem. Talk to us about a Claude Code rollout, or see how we run engagements first.
