Khan Academy and Anthropic are implementing workflows that place teachers firmly in control of AI-generated classroom materials, signaling a broader shift away from fully automated content delivery. Khan Academy's redesigned practice tool requires teacher review and approval before AI-generated questions reach students, while Anthropic's Claude for Teachers adds structured teaching skills grounded in learning science to support lesson preparation and assessment design.

What New Features Are Khan Academy and Anthropic Rolling Out?

Khan Academy is launching updated features in Khanmigo, its AI tutoring assistant, following a six-month collaboration with Google that involved six Google engineers working directly with Khan Academy's product team through a Google.org Fellowship. The updates include interactive visual tutoring for students and a rebuilt practice tool that gives teachers control over AI-generated assignments.

For students, Khanmigo can now generate interactive diagrams when working through math and science problems. Instead of text-only explanations, the AI tutor produces charts, geometric shapes, and coordinate plots that students can manipulate directly. Moving a vertex or dragging a line segment prompts the AI to recognize the change and adapt its response. Early results show signs of increased student engagement and improvements in what Khan Academy calls "next-item correctness," a measure of whether students can correctly answer the next problem on the same skill without AI assistance.

The second major change tackles teacher control. Khan Academy rebuilt its Practice My Knowledge tool to allow educators to use Gemini, Google's AI model, to draft multiple-choice questions, upload their own materials as reference, and then review, edit, or reject questions before releasing assignments to students. Teachers can set work as scored or unscored and receive structured performance reports once students complete it.

Anthropic and Learning Commons are taking a different approach with Claude for Teachers, adding two new teaching skills grounded in learning science: Lesson Preparation and Check for Understanding. The Lesson Preparation skill helps teachers work through an upcoming lesson before teaching it, including identifying key instructional moments and anticipating where students may struggle. The Check for Understanding skill generates short math assessments, typically one to three items, designed to surface what students understand about a particular standard and inform what teachers should teach next.

How Do These Tools Differ in Their Approach to Teacher Control?

Khan Academy's Practice My Knowledge implements an explicit teacher-approval workflow: Gemini generates draft questions, but the educator reviews, edits, or rejects them before students ever see the material. This places the final decision-making power directly with the teacher.

Claude for Teachers takes a different path. Learning Commons describes its Agent Skills as packages containing instructions, references, and guardrails that help AI assistants complete particular K-12 teacher workflows more consistently. Rather than generating assignments for approval, these skills assist teachers in their own preparation and assessment design work. The Lesson Preparation skill helps teachers internalize existing lessons through conversation, while Check for Understanding generates focused assessments to inform instructional decisions.

"Google is a tremendous partner. The Google.org Fellows have supercharged our engineering effort, allowing us to produce several new AI-powered features in time for students and teachers to use these tools in the new school year," said Sal Khan, founder and CEO of Khan Academy.

Sal Khan, Founder and CEO of Khan Academy

Learning Commons has made its Agent Skills cross-platform and model-agnostic, rather than restricting them to Claude. The open skills can be installed in agent environments that support the format, with Learning Commons listing Claude Code, Cursor, and Codex as examples. Developers can also use the skills and its Model Context Protocol (MCP) server through application programming interfaces (APIs) when building them into other products.

How Teachers Can Use These New AI Tools Effectively

• Review and Approve Before Release: Use Khan Academy's Practice My Knowledge tool to examine all AI-generated practice questions, edit them for accuracy and alignment with your curriculum, and reject any that don't meet your standards before students access them.
• Prepare Lessons Strategically: Work through upcoming lessons with Claude for Teachers' Lesson Preparation skill to identify key instructional moments, anticipate student confusion points, and prepare targeted responses before teaching, rather than relying on AI to create entirely new lessons.
• Use Focused Assessments for Instructional Decisions: Generate short math checks focused on specific standards using Check for Understanding to identify what students currently understand, then use that data to determine what to teach next rather than following a predetermined sequence.
• Monitor Visual Learning Effectiveness: When students work with Khanmigo's AI-generated diagrams and interactive visuals, observe which types of visual supports are most effective for your students and provide feedback to refine future outputs.

"We are incredibly proud to support Khan Academy's visionary approach to education. By combining their deep expertise in student learning with Google's AI engineering, we are empowering teachers and students with tools designed to make learning more effective, engaging, and accessible than ever before," said Jen Carter, Head of Technology at Google.org.

Jen Carter, Head of Technology at Google.org

Sandra Liu Huang, President of Learning Commons, stated that the two organizations built the skills "grounded in learning science and designed to support high-quality instruction". Learning Commons emphasizes that its skills draw on its Knowledge Graph and are intended to produce materials aligned with pedagogy and academic standards.

Liu Huang, President of Learning Commons

Learning Commons does not guarantee that outputs will be accurate or suitable in every context. Its repository explicitly states that resources are provided "as-is" and that users remain responsible for evaluating whether generated materials are appropriate for their particular use case.

The new features are available now to districts using Khanmigo, and Anthropic is expanding Claude for Teachers as a free enterprise offering for schools and districts, moving the product beyond individual teacher use while adding more structured instructional workflows. This shift reflects a broader recognition that effective classroom AI requires human judgment at critical decision points, not just automation for its own sake.