Implementation trends and considerations
We saw the same product—even the same activity—land differently in different classrooms depending on the way the activity was introduced, what the teacher did while students were using the product, and how the tool fit into the plan for the rest of the class.
Implementation matters to the experience and impact of a tool as much as its design. Supporting implementation of AI-powered products will require attention to both factors that have always mattered in classroom implementation and some new questions.
The strongest implementation examples of classroom use we saw were in schools where the leaders created clear expectations for use and offered support for using the product.
The conditions that matter across all products include:
- Clear points in the scope and sequence where the product should be used.
- Time held in the classroom schedule for its use.
- Predictable routines around product use (e.g., how the teacher frames purpose and transition, expectations for engaging solo or with peers).
- Login, logoff, and device logistics worked out in advance.
- A clear vision for teacher actions during product use.
Where those conditions were missing, even products with clear uses weren't used as intended or were used inconsistently.
Students engage more productively when teachers frame tool use in service of learning.
We saw teachers frame the purpose of the product in three ways: 1) completion (e.g., "we have to get our ABC product points or minutes"); 2) practice (e.g., "we are using ABC to complete our practice test"); and 3) learning (e.g., "we are working on writing stronger claims, and we are going to use ABC tool to get a first round of practice writing claims").
Students benefit when teachers actively monitor engagement and learning progress and provide real-time support to students who are off track.
When teachers used a tool's dashboard to refocus students who were off task, address misconceptions, pull small groups, or pause activities and launch a whole-group discussion, students were more engaged and, we think, learned more than when teachers reviewed the dashboard after the fact or not at all. We did not see any examples of products or uses that could support strong learning without an active teacher role.
Coaches, assistant principals, and principals matter immensely to how teachers use a product, but few products are really designed to support these groups.
We saw a wide range of leader involvement in tool use. In some classrooms, the leader did not know the teacher was even using the tool. In others, the leader knew it was being used but did not know much about it or what it should look like. In a few, the leader had a clear understanding of the product, how it should be used, and what they wanted teachers to do while students were engaging with it.
Where the leader had a strong understanding and supported teachers in using the tool, we saw teachers taking desired actions more frequently. With an admittedly small sample that we could observe in action, we did observe that the products that gave leaders actionable insights tended to be implemented more consistently in line with that vision. Where adoption was left to individual teachers, use varied more from classroom to classroom.
Effective implementation of some AI-powered products and uses is easier to support than paper-pencil instructional materials or other education technologies.
When products provide helpful feedback to students directly, instruction is more focused and doable for teachers, which also makes supporting key actions easier for leaders.
Because AI-powered products can dynamically respond to student work, products have the ability to provide students with the first or multiple rounds of feedback such that students can progress in their learning without being solely dependent on the teacher.
AI-powered products can nudge key actions for teachers and leaders in real time.
AI-powered products can support real-time analysis. They can productively nudge teacher and leader actions, driven by real patterns in student work—not categories of hypothetical options as previous edtech did. The instructional setup of many AI-powered products lends itself to helping teachers make immediate adjustments.
AI-powered products and uses create some new implementation challenges and considerations.
More than with prior edtech products, teachers have to explain how students should think about the AI role relative to the teacher role.
In some classrooms we observed, the teacher positioned meeting the expectations of the tool as the goal (e.g., “the goal is to get 10 questions right as measured by the tool”); in others, the tool was positioned as an assistant in a goal that the teacher owned (e.g., “my goal is for each of you to succeed in two-step algebraic problems, and I’ll be watching your work with ABC tool to see how you are doing towards that goal, but let me know if you think the AI gets it wrong—my vote is the one that counts”). Because AI interacts with students more dynamically and personally than prior education technology, and because students have a range of feelings about AI in their learning, these are often new conversations for teachers that require support.
Leaders face new questions about how to use the flexibility and adaptability that many products offer.
For example, when a school or system leader considers a writing assignment on an AI-powered tool, they will need to make some decisions:
- Integration: For products used in core instruction, what part of the curriculum does this activity fit into? Does it substitute for a different activity or is it added?
- Rubric: Do we use a standard rubric or can teachers vary it? Will we use the rubric from the curriculum or the rubric from the state test?
- Grading: Will the AI feedback or evaluation of student work be used as a grade? What do teachers need to do to confirm accuracy before grading? How do we treat the data that comes out of this product relative to the data from interim assessments?
- Coaching: Will we conduct teacher evaluations while using the product? How do we factor the AI interaction with students into the evaluation? What would change about what we are looking for in the lesson?
Different types of products demand different kinds of training for teachers.
Teachers told us that getting from "I know the tool exists" to "I know how to use the tool well" took sustained support, not a single rollout session. Teachers who received coaching on tools said feedback helped them stay current and use them better.
- Teachers using LLMs personally or with students need support to spot hallucinations, prompt well, and understand new product features.
- Teachers using education-focused, multipurpose platforms need to understand how to navigate the product and school system-specific restrictions, as well as support in thinking through how to integrate the platform with other elements of the system’s academic strategy.
- Teachers using targeted instructional tools need to understand when and how the product should be used in conjunction with the other elements of the academic system and a mechanism to learn about product updates.
Products can change faster than schools can schedule professional development time.
The rate at which products are evolving provides an unprecedented opportunity to shrink the time between identifying a better way to do something and giving many more students access to that improved experience. For example, when a product developer identifies a better approach to feedback on a question, they can fix it for thousands of students immediately, rather than having to go through a series of professional learning loops to apply the insight at scale. However, this also creates an unprecedented requirement for teachers and leaders to track and understand how features are changing.