How Stanford’s CRAFT Is Rewiring AI Literacy Through Teacher Co-Design

“Incentives in the private sector favor promoting their tools in educational programs, when there is a large and complex ecosystem of tools that will come and go – so keeping AI literacy focused on enduring and durable ideas and practices that are not bound to specific tools is important.”

Victor R. Lee, PhD, Faculty Lead for the Stanford Accelerator for Learning’s Initiative on AI and Education

Schools are introducing artificial intelligence faster than they are deciding how it should be taught. Districts approve platforms, teachers experiment with generative tools, and students quickly learn how to prompt systems for answers. Most public debate centers on access, guardrails, and technical capability. Far less attention is given to who defines what students are meant to understand about artificial intelligence in the first place.

AI literacy is not created by exposure alone. Students do not learn how models are trained, how outputs can mislead, or how data choices shape results simply by using a tool. Those ideas emerge through instruction. Curriculum determines whether artificial intelligence is treated as a shortcut for assignments or as a human-designed system shaped by tradeoffs, bias, and uncertainty. When that curriculum is built outside the classroom, teachers are positioned as implementers rather than authors.

Stanford’s CRAFT initiative intervenes at that structural layer. Developed through collaboration between Stanford students and practicing high school educators across the United States, CRAFT produces free, research-informed AI lesson materials grounded in classroom realities. Instead of distributing polished, static content, the program treats teachers as curriculum co-designers. Their experience managing time constraints, varied learning goals, and student backgrounds informs the design from the beginning. Researchers contribute learning science and technical insight, but authority over what works in practice is shared.

This shift reframes AI education. Rather than centering tool adoption, CRAFT focuses on making artificial intelligence legible, pedagogically meaningful, and adaptable across contexts. The initiative operates on the premise that equitable AI literacy requires more than access to platforms; it requires curricula shaped by the educators who understand how students actually learn.

To understand how shifting curriculum authority toward practicing educators changes what students learn about artificial intelligence and how that knowledge persists in classrooms, we speak with Victor Lee, Associate Professor at Stanford University’s Graduate School of Education and Faculty Lead for AI and Education at the Stanford Accelerator for Learning.

Dr. Victor R. Lee is an associate professor in the Graduate School of Education at Stanford University and serves as faculty lead for the Stanford Accelerator for Learning’s initiative on AI and Education.

He specializes in academic resources for data literacy, K-12 data science education, and AI literacy for students and educators.
Dr. Lee’s book, Learning Technologies and the Body (published by Routledge), is the first compendium of contemporary research on embodied technologies for learning. With Abigail Phillips, he published Reconceptualizing Libraries: Perspectives from the Information and Learning Sciences (2018). His most recent book, released in 2025, is Advancing Data Science Education in K-12. He is a past-president and fellow of the International Society of the Learning Sciences.

Teacher co-design changes how the curriculum is built and how it functions in a classroom. Instead of arriving as a completed unit, the material is written so it can be revised, shortened, or reorganized during instruction. In AI education, this matters because classes differ in schedule, student preparation, and course objectives.

“Teachers adapt for their contexts, so making sure materials are adaptable (such as Google Docs rather than PDFs) is important—others tend to focus on finished form when a segment of teachers want things to tweak and make their own,” Dr. Lee explains.

The format signals how a lesson is meant to be used. An editable document allows a teacher to adjust terminology, add examples, or remove sections to match the class level. The lesson develops inside the classroom rather than remaining fixed beforehand.

Teachers also look for elements they can incorporate into existing lessons. “For the things that are polished, they want stuff that can be extracted out—such as pre-existing videos or mini-apps, but then want to fit into their own lesson structure and tempo.” A short video can anchor a discussion on training data, while a small interactive activity can accompany a media literacy assignment. The components support instruction already in place.

Learning goals vary across courses. As Dr. Lee puts it, “There is a lot of variety in learning goals also for teachers—so what they want students to get out of an experience related to AI is highly variable.” Some teachers focus on how models generate responses, others emphasize evaluation of automated outputs, and others connect AI to ethics or civic reasoning. The materials accommodate these different purposes by remaining adjustable.

CRAFT organizes its lessons so teachers select and revise the pieces they need and align them with local standards and classroom pacing. The curriculum operates as a framework that educators actively shape during teaching.

Debates about artificial intelligence in schools often focus on access to devices and software. Whether students can log into a platform matters, but access alone does not determine understanding. What shapes understanding is instruction. Students learn how systems behave, where they fail, and how to evaluate outputs only when those ideas are taught in context.

Dr. Lee explains that effective use depends on interpretation as much as availability. “There is a lot to know about whether and how to use the tools effectively and to ensure that the value for what teachers and students care about is there.” Teachers need ways to connect technical behavior to classroom learning goals. Lessons must help students see why an answer appears, not just how to generate one.

He also points to a familiar gap in the sector. “There is sort of a joke in the field that ed tech as a sector provides solutions that teachers never asked for.” Tools often arrive built around assumed needs rather than classroom realities. Teachers then spend time adapting them or decide they do not fit the course.

Co-design begins from classroom priorities instead. Dr. Lee describes the purpose directly: “So, engaging teachers in the curriculum materials creation process and fitting it to the concerns they have with other existing curriculum or content or relevance to students is one important way to address that. CRAFT provides one example of how we can do that.” When educators help build lessons, AI concepts can align with writing instruction, scientific reasoning, or media literacy without displacing existing objectives.

In that sense, curriculum becomes part of equity. Students gain durable knowledge when instruction explains systems rather than simply introducing platforms. The central question shifts from whether students can access a tool to whether they can evaluate and interpret the tool’s output.

Giving teachers a role in designing curriculum introduces practical constraints. Schools operate on tight schedules, and instructional time is already committed to required subjects, grading, and administrative work. Dr. Lee notes the immediate barrier: “Teachers are very pressed for time, so putting in the time commitment and work that something like CRAFT asks can be hard.” Participation requires preparation, meetings, and revision work that fall outside normal classroom hours.

Familiarity with artificial intelligence also varies. Many educators are experts in their disciplines but have limited exposure to machine learning systems. He observes that the field’s novelty affects participation. “AI also can feel really overwhelming and new for folks working in schools who have different expertise, so part of the work involved in developing AI education from an educator perspective is changing educators’ views about whether they can participate in an activity like this.” Co-design, therefore, includes both confidence-building and lesson development.

Compensation becomes another structural issue. Dr. Lee explains, “We also see it as very important to compensate teachers for their time and expertise, but finding funds for that can always be challenging, especially as government funding for education research and development has been retracting.” Without funding, participation depends on voluntary effort, which limits how many educators can realistically contribute.

Teachers also face financial incentives elsewhere. “Some teachers are incentivized to try to commercialize their work through teachers pay teachers, in part because they are not compensated enough for the work they already do.” Independent marketplaces provide income opportunities that formal programs often cannot match. Recognition within school systems also matters. He adds, “It would be nice also for districts to recognize and reward teachers for their participation in these activities, socially and when possible, financially, and create space in their work schedules.”

Institutional structures present additional friction. “There is also huge variability in forms of co-design, so we probably need more progress in understanding the advantages and trade-offs for different forms of co-design,” Dr. Lee says, noting that schools and researchers still lack shared models. Universities also evaluate outputs differently. “This also doesn’t tend to yield the outputs that are valued at universities because it doesn’t fit traditional research output canons.” Work that improves classroom instruction may not translate into publications or grants.

These constraints explain why teacher co-design remains uncommon despite interest. The work requires time, funding, and recognition across school districts and research institutions.

The pace of development in artificial intelligence makes curriculum decisions consequential. Research labs and companies advance capabilities quickly, while classrooms operate on yearly cycles and established learning standards. The distance between those environments shapes what students are asked to learn.

Dr. Lee warns that the gap can widen if instruction follows only the frontier of technical possibility. “There can be a real disconnect with what is happening and being explored on the edges of possibility in research and industry with what the vast majority of people are trying to work on with AI, so I fear that gap widening and creating unrealistic expectations for what students need to learn or how teachers teach.” Students encounter systems in everyday contexts such as writing, searching, and media consumption. Instruction, therefore, needs to prepare them to interpret ordinary uses rather than mirror specialized research settings.

Commercial incentives further influence classroom exposure. He notes that platform providers naturally promote their own systems in educational settings.

“I also think that incentives in the private sector favor promoting their tools in educational programs, when there is a large and complex ecosystem of tools that will come and go – so keeping AI literacy focused on enduring and durable ideas and practices that are not bound to specific tools is important.” Tools change quickly, but the underlying concepts of training data, uncertainty, and evaluation remain relevant.

A curriculum built through teacher participation supports that durability. Lessons emphasize reasoning about outputs, questioning sources, and understanding system behavior, rather than mastering a single interface. Students learn practices they can apply across platforms and over time.

Ultimately, the result is a different definition of literacy. AI education becomes preparation for judgment rather than familiarity with a product. And as technologies evolve, the knowledge students carry forward depends on whether they were taught to use systems or to understand them.