#104: Reflections on PD Design (2026)

It is quite easy for workshops to fall into a familiar cycle: introduce a feature, create an exemplar, give teachers some hands-on time, and repeat the process when the next feature arrives. What I've come to appreciate more deeply is that different groups need different workshop experiences.

One of the earlier sessions I supported this year was for a group of Pre-U Geography teachers. As SLS was less commonly used by this group, we developed a comprehensive fieldwork SLS package on the topic of community response to climate change for participants to experience. The package was designed to guide learners through the full geographical inquiry process, from activating learning about community responses to climate change, to pre-fieldwork planning, actual fieldwork, and post-fieldwork analysis and evaluation. The package also includes additional resources to extend students’ understanding of community-led climate action in Singapore. Team Activities support collaborative planning, data collection and analysis, while the Short Answer Feedback Assistant provides immediate feedback to help students refine their thinking as they progress through the fieldwork.

Outline of SLS module co-created with AST and CPDD GYU:

We got teachers to experience one part of the package as learners, focusing on how SLS could scaffold students’ observations before they moved into fieldwork. One scaffold was the 8Way Thinking template, which responded to a very practical challenge: asking students to “observe” does not necessarily mean they know what to notice, or how those observations might eventually lead to geographical questions worth investigating.

Part of the SLS module that teacher participants experienced:

Through this, teachers could see how SLS was used not simply as a collection of individual features, but to connect and scaffold the different stages of the learning experience, with AI-enabled feedback providing students with more immediate feedback along the way.

Teacher's feedback:

Teachers surfaced different possibilities for the SLS exemplar, including its potential to support self-directed learning, as well as ways in which they might adapt parts of the package for their own fieldwork contexts or apply a similar approach to other fieldwork topics. Ultimately, the goal of developing the exemplar was not for teachers to use it wholesale, but to give them something concrete to experience, critique and take apart, so that they can then decide what might work in their own contexts.

For another workshop on how teachers can better support students in tackling Geography evaluative essays, we explored how Short Answer Feedback Assistant (ShortAnsFA) can use a rubric to evaluate students’ responses and provide feedback. More importantly, we wanted teachers to examine how the rubric itself shapes what the AI pays attention to and the kind of feedback it gives students.

Rather than spending precious workshop time getting teachers to log in to SLS and configure ShortAnsFA, we distributed physical handouts containing essays from Monica's [LT/Geography at Bowen Sec] students, alongside her feedback and two versions of feedback generated by ShortAnsFA.

Part of the physical handout used for the session:

Teachers compared what Monica and AI noticed, the tone and specificity of their feedback, what Monica picked up from the student’s thinking that the AI did not, and what students could act on. They also compared feedback generated by ShortAnsFA using different rubrics.

One of these was an adapted SOLO Taxonomy rubric that students had previously been introduced to. It made the qualities of a good argument, example and evaluation more explicit: progression was not simply about writing more, but about moving from listing towards linking ideas, from giving examples towards using them to support reasoning, and from stating a view towards weighing evidence using appropriate criteria and context.

When these same qualities were used to guide ShortAnsFA, its feedback was more closely aligned with what students had already been taught about quality. Comparing this with feedback generated using a more conventional assessment rubric also helped teachers see how different rubrics direct the AI’s attention differently, and therefore create different opportunities for students to improve their work.

This reinforced an important point: the quality of AI-generated feedback does not begin with the AI. What we want students to learn, how we articulate quality through the rubric, and the instructions we give the AI all shape the feedback that eventually reaches the student.

During the discussion, one teacher pointed out that the feedback was probably too long for weaker students. This gave me an opportunity to share that teachers could now provide custom instructions to ShortAnsFA, including instructions on how the feedback should be structured and the ideal length. This exchange was particularly useful because it addressed something I have noticed with teachers' experience with SLS features: They may have tried a feature when it was first released, decided that it did not perform to their expectations, and never returned to it. Meanwhile, the feature may have changed considerably, and simply announcing an enhancement may not change that earlier impression. Encountering it again in relation to an actual problem probably has a better chance of doing so.

Importantly, by removing the friction of navigating the SLS platform and setting up the feature, we could focus first on what ShortAnsFA made possible. Teachers saw its value, which gave them a reason to figure out how to set it up afterwards, rather than spending workshop time learning the mechanics before seeing what the feature could offer.

There was also a deliberate decision about what not to include in the workshop. We had initially considered sharing about Annotated Feedback Assistant (AFA), but did not think it was working consistently enough yet for what we wanted participants to experience. Bearing in mind that introducing more features does not necessarily make for a better workshop—in fact a poor first experience can easily become the impression a teacher carries away with them—we stuck with ShortAnsFA, since it gave us a more reliable way to have the conversation we wanted about feedback.

Side note; I also took the chance to demonstrate that ShortAnsFA could read students’ handwritten work 😉. (For the workshop itself, I separately converted the handwriting to text and included it on the handout simply to make the essays easier for teachers to read and analyse.)

With student teachers, I took yet another approach. Many of the Geography PGDE trainees had limited experience using SLS themselves and tended to see it mainly as a platform for asynchronous learning. Some were also concerned about its complexity and how it might work when students have different levels of readiness.

For their workshop, I facilitated a synchronous SLS lesson, with the NIE trainees taking on the role of students. This allowed them to experience how SLS could support a synchronous lesson, including how a teacher could design and customise the learning experience, monitor students’ progress, and provide timely feedback.

Part of the SLS module that the NIE trainees experienced:

Feedback:

The three sessions were quite different, but that is probably the point. Not every workshop needs to be hands-on in the same way. Sometimes it is useful to experience the whole lesson, sometimes participants need to use the platform themselves, and sometimes it is better to remove the platform from the foreground altogether so that the discussion can concentrate on the student work and the learning.

Tl;dr, key consideration:

Another realisation was that there is already a lot that can be learnt by taking an existing lesson and looking at how it might be adapted; there is no need to create new resources for every workshop.

In a workshop on formative assessment and 21CC, for example, we worked with an existing lesson on urban hazards. A task in the original lesson asked students to identify hazards found in urban environments. A relatively small change to the question could make students' geographical thinking visible. Beyond identifying hazards, students could be asked which hazards they thought were most common in Singapore's urban neighbourhoods and to explain why. This will require them to consider population density, human activities and the characteristics of the urban environment.

Likewise, asking students whether air pollution in Singapore might worsen or improve can go beyond a simple prediction if they have to consider Singapore as a place, as well as processes operating at other scales, such as regional haze.

The changes we were making were relatively small, but they shifted the task from naming hazards to explaining patterns; from describing impacts to understanding causes; and from making local observations to reasoning across different scales.

Adapting an existing SLS lesson to strengthen formative assessment:

We then looked at which 21CC might be a natural fit for the lesson. This was important because the intention was not to fit as many competencies as possible into a lesson, especially since different subjects and topics lend themselves to different competencies. Teachers then examined the E21CC developmental milestones before returning to the lesson and considering how the learning experience might support the development/progression of the competencies.

A similar approach shaped the EZ Combined Cluster Humanities Workshop on ipsative assessment. We adapted an existing Social Studies SLS Community Gallery module to highlight elements of an ipsative learning experience.

Participants experienced the lesson organised around: Data → Feedback → Iteration → Reflection. We then unpacked what made the experience ipsative and how each part of the process contributed to students becoming more aware of their own progress.

Part of the SLS module that teacher participants experienced:

The same module can be revisited through a different lens—formative assessment, conceptual understanding, 21CC, differentiated instruction, etc, and often there is more value in showing how an existing resource can be adapted than in continually giving teachers another finished resource.

Tl;dr, key consideration:

Another change in our later workshops was where the exemplars were coming from. Our teacher leaders were increasingly bringing in lessons, resources and AI experiments that they had developed themselves. At our first Senior Teacher / Lead Teacher Professional Learning Session (ST/LT1), we adopted a World Café format, where teacher leaders shared different ways in which they had been experimenting with AI for T&L. The examples ranged from concept-based inquiry and reflective inquiry to simulations and scenario-based learning.

Participants moved between the examples and listened not only for what had been created, but for the pedagogical considerations behind the design. Peishi [ST/Geography at Naval Base Sec] shared her experience developing a sustainable housing simulation with AI. What was useful was not simply the final simulation, but the articulation of the thinking process behind the iterations.

Sharing of the design & snapshots from the simulation:

For one of the scenarios in the simulation, the AI initially suggested Tengah as the setting. Peishi changed it to Lorong Chencharu because it was beside her students' school and the students could actually see the clearing and construction taking place. She also replaced some of the content suggested by AI with concepts and examples that better reflected the geographical terminology and learning her students needed, including the urban heat island effect and the use of green roofs and walls.

Prompt-development process:

The changes were possible because Peishi knew her students and knew the Geography she wanted them to learn. AI could help with developing the simulation, but it could not replace that knowledge. The final product is only part of the story. The iterations—what the teacher accepted, rejected and changed, and why—are often more useful for other teachers to see.

The session also showed participants the possibilities of what teacher sharing within their department or school can look like. Instead of presenting a finished resource as something for colleagues to replicate, the resource becomes something that everyone can examine and critique and refine.

There is a parallel opportunity with students. At the S1 Cluster Humanities Workshop, one of our discussions centred on AI literacy and the role teachers can play in modelling what critical use of AI looks like. One possibility I shared was to make some of our own use of AI visible to students. If it is useful for teachers to see another teacher’s AI process—not just the final product, but what the AI generated, what was accepted or rejected, what was changed and why—why not let students see some of that process too?

A teacher could, for example, show students an AI-generated resource they had used in class and unpack how they worked with it: what they prompted the AI with, what came back, what did not quite work, and what they changed because of their knowledge of Geography, their students and the intended learning. Students then get to see what critical use of AI actually looks like in practice, rather than only being told that AI outputs need to be checked and evaluated.

Tl;dr, key consideration:

As more teacher leaders started experimenting with AI-enabled interactives and simulations, we also wanted to be deliberate about how we framed their use of AI. The aim was not simply to encourage teachers to use AI to create more things, but to keep bringing the conversation back to purpose: does this need to be an interactive in the first place, and does AI need to be involved?

This felt increasingly important as AI made it much easier to create things that would previously have required considerably more time or technical expertise. An interactive or simulation can now be produced without knowing how to code. However, the falling cost of creating one should not lower the bar for deciding whether it is pedagogically necessary. If anything, as AI makes more things possible, professional judgement matters more.

At ST/LT2, a simple question was posed: "Must we use AI solutions in teaching?"

The perspective shared by Yanjie [SH/Geography at CCHS (Yishun)] considered both the complexity of what students were trying to understand and the amount of interactivity or immediate feedback that was needed.

Some things can be understood perfectly well through direct teaching or reading. Some may benefit from a thought experiment or discussion. Others could work as a card or board game without technology. There are then situations where a technology-mediated simulation is useful because students need to manipulate variables and quickly see the consequences of their decisions. There is also a point where too many variables simply become frustrating.

At the same session, participants examined interactives that other Geography teachers had developed and discussed them from the perspective of the intended learning. For one simulation on alternative energy involving local energy stakeholders such as SP Group and the Energy Market Authority, participants appreciated that the scenario was grounded in a recognisable real-world context. At the same time, they questioned whether the number of variables and possible permutations might overwhelm students, and whether the experience would lead to sufficiently deep learning.

For another interactive on seafloor spreading, the discussion returned to the learner. Participants considered what prior knowledge needed to be activated, whether the sequence provided enough scaffolding, which variables students should attend to first, and how the experience might need to differ for students of different readiness levels. They also considered how learning data could subsequently be used to consolidate understanding and address misconceptions.

These are the kinds of conversations I hope we continue having. As AI makes more things possible, being able to create something is increasingly the easy part. The harder—and more important—part is deciding whether what we create actually helps students learn.

The fact that we can build something now is not, by itself, a reason to build it. Sometimes making it simpler—or not using technology at all—may be the better design decision.

Tl;dr, key consideration:

Another thing I appreciated about the later sessions was that teacher presenters were not only sharing polished examples of "successful" SLS use. At one of the workshops, teachers spoke candidly about SLS features they had tried where the outcome was not what they expected. One example concerned the feedback produced by AFA.

These conversations are useful because there can be several reasons why something does not work as expected. It might be how the activity was designed. It might be how the feature was configured. Another teacher may have found a different way of using it, or the feature itself may simply still have room for improvement.

When it was the latter, we were open about it. We acknowledged where there were limitations, shared that the feature was still being improved, and took the issues raised back to the team. I think that openness matters for building trust. If we want teachers to share candidly what happens when they use SLS with their students, then our professional learning spaces also need to be places where they can surface what did not work without us feeling the need to defend the technology.

It also makes the relationship less one-directional. We may be sharing new SLS affordances with teachers, but teachers who are actually using these affordances with students are giving us useful feedback about how they work in practice. That feedback is difficult to get if the features only ever appear in demonstrations.

Tl;dr, key consideration:

There is one important prerequisite for much of what I have described so far: that our teacher leaders are using SLS in their teaching. As their familiarity with and use of SLS has grown over the last two years, the nature of what we can do together has changed too.

Increasingly, we start with an existing SLS module and think about how it might be adapted rather than always creating something new. Our teacher leaders try different approaches with their own students and bring those experiences back to the Subject Chapter. They share resources they have designed themselves, including the thinking and iterations behind them, and because they are actually using SLS in their classrooms, we can talk about what worked, what did not, and what their students actually did.

The North Zone Learning Series on data-driven thinking was a good example. Doris [ST/Geography at Ahmad Ibrahim Sec] had first enacted an SLS lesson with her own students. Workshop participants were subsequently added as observers to the class group, which allowed them to work with the students' actual responses rather than responses prepared for a demonstration.

The session started from a fairly familiar use of learning data: looking across student responses, identifying possible gaps and considering what might need to happen next. Depending on what the data showed, that could mean an intervention for the whole class, a particular group of students or an individual student.

One task asked students to refer to a live Singapore traffic map, decide whether there was traffic congestion and justify their conclusion with evidence. The responses varied. Some students gave a clear conclusion supported by relevant evidence from the map, while others provided limited or less relevant evidence.

We could have treated this simply as information about what the students understood and what intervention they might need. Instead, we also looked at what the responses might tell us about the question itself: could the way the task was phrased have contributed to some of the responses we were seeing?

We then used Data Assistant (DAT) to analyse the responses and also asked it to suggest how the question might be redesigned for greater clarity. It surfaced issues such as the scope of the question, the broadness of asking students “Why do you say so?”, and the ambiguity around what counted as evidence.

DAT gave us several suggestions for how the question might be improved, but we did not take all of them up. Ultimately, the teacher still had to decide what made sense for the intended learning. The value of DAT was in offering another perspective on the task, rather than making the design decisions for us.

Use of SLS DAT to inform task design:

There is also a more basic point here: we could only have this conversation because the lesson had first been enacted on SLS. Everyone could easily refer to the student responses captured, use DAT to help surface patterns, and consider what those patterns might tell us about both the students' learning and the design of the task. Without that classroom use, we would have had to create sample responses for the workshop or discuss the possibilities in the abstract. The workshop discussion would have been quite different, and likely not as rich.

Having the lesson on SLS also made it easier for the practice to travel beyond that particular classroom. If other teachers found the lesson useful, they can easily make a copy and adapt it for their own students. This would have been less straightforward if the lesson was spread across different platforms/tools.

So while getting teachers to use SLS is not the outcome we are ultimately interested in, it isn't something to gloss over either. That actual use provides the conditions for classroom practice to become something that others can examine, learn from and build on.

Tl;dr, key consideration: