AI for teacher development: learning alone or with your team

Generative AI has been inside Chilean staffrooms for two years. According to TALIS 2024, 55 percent of Chilean teachers already use it at work — nearly twenty points above the OECD average — and 68 percent of those who do use it for lesson planning. The adoption is already happening. The question facing a principal or head of curriculum today is not whether their teachers will use AI, but under what conditions, and what that leaves behind.

The conditions matter because using AI alone and using it inside a community of practice do not produce the same result. Teachers who adopt these tools individually tend to stay in low-cognitive-demand uses — generating worksheets, producing materials, planning from scratch — without developing any shared judgment about when AI outputs suit their specific class and when they do not. AI tools for teacher development are more effective when they are embedded in communities of practice. This article explains why, and what a school leader can do this week.

Why individual use is not enough

Kennedy (2016), reviewing 30 years of research on professional development, documents a consistent pattern: when teachers adopt technology without shared practice structures, they tend to stay in low-cognitive-demand uses — generating worksheets, producing materials — and rarely develop deep pedagogical reflection on what that technology does well and what it does not. With generative AI, that pattern is amplified. The tool is good enough to produce plausible materials, which makes it easy to adopt superficially and difficult to evaluate critically without peers who push back on your assumptions.

The risk is not that AI produces bad materials. It is that the teacher using it alone has no one to compare notes with — what did you ask, what did you get, what did you decide to do with it? Without that conversation, what remains is efficiency without reflection.

What changes when it is used inside a community

Arefiana, Çomoğlu and Dikilitaş (2024) studied eight English-language teachers in a collaborative reflective practice programme based on ChatGPT, analysed through Wenger's community of practice lens (DOI: 10.1080/17501229.2024.2412769). The central finding: individual ChatGPT use functioned as preparation that deepened contributions to the group. Teachers arrived at sessions with concrete observations — what the AI had said, what had worked and what had not. The group turned those observations into shared pedagogical judgment.

This is not a workshop where everyone listens to the same thing. The cycle of individual use → collective discussion → improved individual use produces learning that neither moment generates on its own. The community does not replace individual use; it turns individual use into material for reflection.

In Chile, the infrastructure for that cycle already exists. The Professional Learning Communities (PLCs) managed by the CPEIP — established under Law 20.903 — are groups of five to eight teachers from the same school, meeting weekly for at least 90 minutes in cycles of peer observation, evidence analysis, and collaborative instructional design. The CPEIP launched a national AI teacher training programme in 2025. The institutional signal already points in this direction.

The objection that deserves an answer

Desimone (2009), in the reference article on effective professional development (DOI: 10.3102/0013189X08331140), warns that studies of professional learning communities typically measure whether teachers "participate" without separately measuring the five features that produce the effect: content focus, active learning, coherence, duration, and collective participation. A PLC without content focus may produce no impact at all. Community does not guarantee the result.

The practical answer for a school leader is this: a community of practice is the structure that makes it most likely that those five features will co-occur and be sustained long enough to produce change. A two-hour AI workshop can have excellent content, but it has neither sustained duration nor sustained collective participation. Community is not the magic recipe; it is the context where the recipe can actually be followed for the time it takes.

Three concrete things you can do this week

Do not run a general AI workshop. Instead, take a PLC or working group that already functions and add a four-week module on pedagogical AI use within its existing protocol. The use case should be specific: written feedback, analysis of recurring error patterns, or differentiated activity design for a particular learning objective. Specificity produces learning; generality produces attendance.

Set the boundaries before you start. Law 21.719, fully in force from December 2026, restricts the processing of personally identifiable student data in external AI systems. The group needs to agree in the first session what can and cannot be submitted to a language model: no data that identifies any specific student. That conversation is not paperwork; it is what transforms AI use from an individual experiment into an institutional practice with shared norms.

Document what the team learns, not just what the AI produces. The output of the PLC + AI cycle is not the generated material; it is the pedagogical judgment the group develops about when and how to use AI in their specific context. Ask that each session end with a brief note: what worked, what did not work, what did we decide not to use and why. That record is the professional capital that accumulates and can be shared with other PLCs in the school.

What we do not yet know

No quasi-experimental study has directly compared AI adoption inside professional learning communities against individual AI adoption, controlling for Desimone's five features. The argument in this article rests on two solid but separate streams of evidence — that well-designed communities of practice improve teaching practice (Vescio, Ross and Adams, 2008) and that without shared practice structures AI tends toward low-cognitive-demand uses (Kennedy, 2016) — but not on a study that connects both variables for the specific case of teacher development with generative AI. There are also no published data on the impact of integrating AI within Chile's CPEIP learning communities.

What we do know is that 55 percent of Chilean teachers already use AI at work, and that AI training is the professional development need they most frequently report. The question is no longer whether adoption will happen. It is what support structure that adoption has when it arrives in your school.