AI in our staffroom: what stuck, what didn't

We rolled out AI tools across a Year 7-12 staff of 92 last year. Free accounts where possible, paid accounts for a smaller pilot group, two PD sessions, an internal Slack channel, and a quiet experiment with what people actually used by Term 4.

Some of it stuck. Most of it did not. Here is the honest tally, written so you can save yourself the same expensive lessons.

What stuck.

Lesson resource generation, but only for specific kinds of resources. Comprehension question sets from a passage. Differentiated worksheets from a single base task. First drafts of email replies to parents. The pattern across the things that stuck is they all answer a discrete, well-defined task that the teacher can scan and approve in under two minutes.

Rubric drafting. Half a dozen English and Humanities teachers built rubrics with AI assistance and then refined them. The AI wasn't writing the rubric. It was producing a starting point that was better than a blank page. About half the staff who tried this kept doing it.

Email triage and template writing. Three of the assistant principals used AI to draft replies to common parent emails. Not to send unedited. To generate a starting point that they then heavily edited. They consistently saved twenty to thirty minutes a day.

What didn't stick.

Lesson planning from scratch. Almost nobody who tried this kept doing it. The output was either too generic to be useful or too specific in the wrong way. The teachers who tried it generally said the same thing: it took longer to fix than to write themselves.

Marking and feedback. Four staff trialled this. None of them continued. The reasons were partly philosophical and partly practical. The output didn't engage with what the student had actually written, only with the surface structure. The feedback took longer to validate than to write fresh.

Image and slide generation. Genuine excitement at the start, very little continued use. The novelty wore off quickly. The slides students remember are the ones their teacher made because the teacher cared, not the ones generated by software.

Student-facing AI tutors. We piloted one with a small Year 9 cohort. The pedagogical results were mixed. The operational results were a clear no. The amount of teacher time required to monitor, correct, and contextualise the AI output was greater than the time saved.

What we learned that we did not expect.

The biggest predictor of whether a staff member kept using AI was not their digital literacy or their age. It was whether they had an existing habit of using digital tools to draft and revise their own work. Staff who routinely drafted emails and resources in a digital tool found AI a small extension of an existing habit. Staff who didn't tended to abandon AI within two or three weeks regardless of how good the introductory PD was.

We have been more careful since then about who we expect to benefit, and more careful about claiming time savings without measuring them.