Claude for Teachers: How Anthropic Is Reimagining AI for Educators with Drew Bent & Sandra Liu Huang

Drew Bent leads Education at Anthropic, where he oversees education initiatives including Claude for Teachers. A former math teacher, Khan Academy engineer, and co-founder of Schoolhouse.world, Drew focuses on applying AI to improve teaching and learning.

Sandra Liu Huang is President of Learning Commons, where she leads the development of open AI datasets and learning science infrastructure for education. She previously held product leadership roles at Google, Facebook, and Quora.

💡 5 Things You'll Learn in This Episode

1. Why Anthropic built Claude for Teachers exclusively for educators.
2. How Learning Commons grounds AI in standards, curriculum, and learning science.
3. The role of open datasets, AI skills, and shared infrastructure in improving educational AI.
4. How Claude integrates with leading edtech tools through new connectors.
5. What's next for Claude for Teachers, including district partnerships and future AI capabilities.

✨ Episode Highlights
[00:03:27] Why Anthropic chose to build AI for teachers first.
[00:05:50] How Learning Commons powers AI with standards and learning progressions.
[00:08:52] Inside Claude for Teachers' integrations with leading edtech platforms.
[00:14:41] Building open AI teaching skills and evaluation frameworks.
[00:17:26] How AI can support meaningful lesson differentiation.
[00:20:54] The vision for open educational AI infrastructure.
[00:22:42] What's next for Claude for Teachers, including district partnerships and research.

😎 Stay updated with Edtech Insiders! 

Follow us on our podcast, newsletter & LinkedIn here.

🎉 Presenting Sponsor/s:

Innovation in preK to gray learning is powered by exceptional people. For over 15 years, EdTech companies of all sizes and stages have trusted HireEducation to find the talent that drives impact. When specific skills and experiences are mission-critical, HireEducation is a partner that delivers. Offering permanent, fractional, and executive recruitment, HireEducation knows the go-to-market talent you need. Learn more at HireEdu.com.

This season of Edtech Insiders is brought to you by Cooley LLP. Cooley is the go-to law firm for education and edtech innovators, offering industry-informed counsel across the 'pre-K to gray' spectrum. With a multidisciplinary approach and a powerful edtech ecosystem, Cooley helps shape the future of education.

Every year, K-12 districts and higher ed institutions spend over half a trillion dollars—but most sales teams miss the signals. Starbridge tracks early signs like board minutes, budget drafts, and strategic plans, then helps you turn them into personalized outreach—fast. Win the deal before it hits the RFP stage. That’s how top edtech teams stay ahead.

Transcript

[00:00:00] Alex Sarlin: Innovation in pre-K to grade learning is powered by exceptional people. For over 15 years, edtech companies of all sizes and stages have trusted Higher Education to find the talent that drives impact. When specific skills and experiences are mission-critical, Higher Education is a partner that delivers.

Offering permanent, fractional, and executive recruitment, Higher Education knows the go-to-market talent you need. Learn more at hireedu.com. That's H-I-R-E-E-D-U.com. 

[00:00:33] Drew Bent: It may sound strange, but we really don't want this learning commons and like grounded in the curriculum and the high quality instructional materials to be a differentiator for Claude for Teachers.

Maybe it is today, but the hope is that the whole field, like no matter which edtech tool you're using or foundation AI model, that if you're using it for education use cases, that you have the context. It shouldn't be hallucinating the standards. It shouldn't be just vibe making the instructional materials.

It should be pulling from high quality open educational resources

[00:01:09] Alex Sarlin: Welcome to Edtech Insiders, the top podcast covering the education technology industry. From funding rounds to impact to AI developments across early childhood, K-12, higher ed, and work, you'll find it all here at Edtech Insiders. 

[00:01:25] Ben Kornell: Remember to subscribe to the pod, check out our newsletter, and also our event calendar.

And to go deeper, check out Edtech Insiders Plus, where you can get premium content, access to our WhatsApp channel, early access to events, and back-channel insights from Alex and Ben. Hope you enjoyed today's pod.

[00:01:49] Alex Sarlin: Welcome to Edtech Insiders. We have an incredibly special episode today. I have a feeling this is gonna be one of our most listened to episodes because we are talking to two amazing edtech leaders who are behind one of the projects that has taken the entire edtech world by storm, Claude for Teachers.

Uh, we are talking to Drew Bent, who leads education as part of Anthropic's beneficial deployments. He also co-founded the tutoring nonprofit Schoolhouse.world with Sal Khan, and prior to that, wrote code at Khan Academy, taught high school math, and has been tutoring students for over a decade. Drew has degrees in physics and computer science from MIT, and an education master's from Stanford.

Not bad at all. Sandra Liu Huang is president of Learning Commons, which funds and builds public AI data sets and resources to help bring more learning science into classrooms. As a philanthropic and technology leader, Sandra advises organizations like Airdif and Play Lab, and previously Sandra held product leadership roles at Quora, Facebook, and Google.

Drew Bent and Sandra Liu Huang, welcome to Edtech Insiders 

[00:02:56] Drew Bent: Thank you for having us. 

[00:02:57] Sandra Liu Huang: Great to be here. 

[00:02:58] Alex Sarlin: It's so great to talk to the both of you. So first, let's just jump into it. Drew, Claude for Teachers is big launch right now. It's built specifically for the educator community rather than students. And the announcement about it points to evidence that this is really where, uh, AI can reliably strengthen instructional practice by going through the humans, the teachers.

Tell us about how that focus on educators took shape, and what you see teacher-facing AI being able to do for the craft of teaching. 

[00:03:27] Drew Bent: That's right. Thank you, Alex, for having us. As you shared, we're very excited to be releasing Claude for Teachers free for every educator in the United States. You know, I'll be the first to say that AI in education is a contentious issue, especially as we think about student-facing tools.

You know, we at Anthropic have our own concerns about AI's potential impact on students' long-term learning, and we think it can be done in a way that makes sense, but it has to be done in a very intentional way. And so our Claude apps are only directly available for adults, and so we focus this launch on educators.

And you know, I can say this as a former teacher myself, you know, I know how important the job of a teacher is, but also how difficult it is. And you know, most teachers go in, they want to spend time in the classroom with their students in small group settings and one-on-one settings working with those students.

And yet a large part of the job of a teacher is spent in evenings, on weekends, preparing lesson plans, trying to differentiate instruction, making assignments and formative assessments. And so that's where we really want Claude for Teachers, to support the teacher so that they can spend their time in the classroom.

And as we were talking to educators, you know, co-designing this with them, it kept coming up that teachers were finding some value out of existing AI tools, but they often felt like it didn't have the context about their state standards, their curriculum, their particular class And so that was sort of an ongoing challenge because again, these AI outputs you can get, like a lesson plan, it can look great on the surface, but as you dig deeper, you can see that there's big gaps in what would make sense for your class.

And so that's why we were so excited to partner with Sandra and the Learning Commons team, and the core part of Claude for Teachers is this Learning Commons data set that brings in, you know, state standards and high-quality instructional materials, which I'm sure Sandra will talk more about. 

[00:05:20] Alex Sarlin: Yeah, that's a perfect handoff.

So Sandra, right, Learning Commons is the infrastructure layer underneath a lot of this personalization that, that, uh, Drew is mentioning. It's-- gives Claude access to academic standards across all 50 states, the learning components, learning progressions beneath them so that, you know, the Claude is not starting from a, from a cold start ever.

It's actually contextualized. Uh, walk us through what sits in that data layer, and why that structure allows a better learning experience for people using Claude with Teachers. 

[00:05:50] Sandra Liu Huang: Yeah. I think you gave a great summary, and, you know, at Learning Commons, our focus, as you've mentioned earlier, is really thinking about how can we bring quality learning science and research into classrooms more easily.

I think as a sector, we don't have a lot of mechanisms to carry and capture what we know about great teaching and learning, and in our work, we're finding new opportunities to create those data sets and tools to make that more possible. And we're doing that with a lot of people across the field, educational experts, learning science researchers, to really piece together a data set that can be public and open and leveraged by the whole ecosystem to really advance educational tools and make them work better for teachers, as the-- as Drew talked about in some of those use cases that Claude for Teachers enables.

Really concretely, inside the Learning Commons knowledge graph, we've got the standards, and state standards are almost like a baseline thing, but it's really helpful to s- to know which standard you're talking about in a specific way. But what's really powerful is we work with educational experts to break those standards into smaller pieces.

We call those learning components, but think of them as the sub-skills that the standard is trying to describe. In addition to that, once we've broken those out, we've added the connections between them, and so that you can actually have access to the learning progression. How do you connect a concept? For example, you know, in sixth grade you might be teaching ratios in math, but that actually relates to, in fifth grade, thinking about the coordinate plane and thinking about lines and slopes.

But it also connects in seventh grade to percentages and sort of other concepts. And inside Learning Commons knowledge graph, we really have those concepts broken out, but also offer the data of what the connections are, and we can talk about how that plays out. But that is a data set that hasn't been open in a way that's so high quality and something that we want to continue to maintain for the broader field.

[00:07:42] Alex Sarlin: 100%. I-- One thing that excites me a lot about both of your approaches to this is that you take a really partnership, co-design heavy approach. You're working a lot with educators. You're working a lot with other educational organizations and nonprofits. You mentioned how Learning Commons has worked with a number of different education nonprofits to build out its learning progressions and how, how the, uh, components work and things like that.

And then Drew, this announcement had a, a whole lot of different amazing partners, including AFT and Teach for America and nine edtech connectors, this is particularly interesting to our audience, from ASSISTments to Diffit to Magic School to Brisk Teaching to TeachFX, CoTeach, and others. And I think this is a piece that maybe people don't yet understand, even though it's so exciting, which is what does a teacher actually experience when they connect one of these tools, these widespread edtech tools inside Claude?

How does it work in practice? Does the partner functionality s- if somebody is a Magic School user, do they get Magic School through Claude, or do they have to have an account with Magic School to use Claude, or can they use Claude to access Magic School? I'd love to hear you unpack how that works, because it's an incredibly exciting sort of team approach to edtech development.

[00:08:52] Drew Bent: We're, we're very excited about these nine edtech connectors, and hopefully it's just the start, um, because there's many more than just, you know, nine edtechs out there. And, you know, ultimately it comes down to how can we make this simplest for the teacher? And as we know, there's many edtech tools out there, and many educators are using different combinations of tools.

And in an ideal world, you want all of these to be connected and interfacing with each other in, in a secure and private way that makes it convenient for the educator And so many of these edtech tools are powered underneath the hood by Claude. And we also think it's important within Claude for Teachers to almost serve as kind of a mission control, where in partnership with these other edtechs, you can bring in their strongest capabilities into Claude for Teachers.

Most of these connectors, they're optional. Not every teacher is gonna have an account with Brisk or Magic School, but if they do, they are able to sign in. I think most of this is through OAuth. And so if they have an account with Brisk Teaching, they're able to connect that through Claude for Teachers.

And what that means is we build Claude for Teachers really on the teacher-oriented workflows. But in c-case of Brisk, they also have the ability for teachers to create assessments, you know, formative assessments for students, put it into, for example, a Google form, and then distribute that and share with their Brisk class, you know, in a safe way.

And so that's really exciting because now a teacher can be using Claude for Teachers, they can sign in if they have a Brisk account, and then they're able to create Brisk assignments for their students that then they can bring back, you know, into the, the interface that students are already using. Um, so that's just one example, but you can-- we can go through all of the different edtechs, and we're excited, uh, to grow this, uh, even further.

[00:10:35] Alex Sarlin: Yeah, and I'm sure our audience is also excited to hear that you're gonna grow it further. I know that many people love the idea of being able to be connected and be able to be, uh, integrated into workflows, especially through an incredibly powerful tool like Claude. And Sandra, building on what Drew just shared, many of these edtech companies and connectors, right, we, we all know a bunch of these folks, they are amazing, and you're directly working with a number of them as well.

Learning Commons is really about an infrastructure that can be used across the AI ecosystem and across the educational ecosystem, both with Anthropic and also directly with edtech providers. Tell us what your partnerships look like with, uh, individual edtech companies, and how the Learning Commons infrastructure supports them in their missions.

[00:11:17] Sandra Liu Huang: Yeah. I, I think as Drew talked about Claude for Teachers as a way to sort of at the teacher interface level kind of bring in their different tools, Learning Commons, again, is in that dataset level, right? It's in the tooling, um, that is trying to really raise the quality and bring more learning science into the data layer that's available to the broader edtech ecosystem.

And so we are partnered with many of those companies that are integrated with Anthropic, and I think hopefully more and more of them. But essentially, you know, the-- our view is that we will continue to work with researchers, with other funders in the education space, with other, uh, nonprofits that bring excellent expertise to really curate and build these datasets in different, in different formats, right?

To create these different tools that can be public goods. And then, you know, we hopefully have this shared corpus as a field that it gives a mechanism for translating great things into classroom practice. And the Learning Commons team, essentially what we do is also put the technical pieces to make it easier for the builders.

Right. And so that is our platform tools. We'll later we'll talk about skills that we've co-developed, the data sets, our evaluation tools. This whole set is essentially built with the field for the field so that whether you are Cloud for Teachers or a, a tech developer with particular use cases that you're working on, you can actually leverage the same bar of quality, of high quality connections, the learning progressions I talked about earlier, and we can all sort of move the field forward because we have this shared common starting point that's continuing to get better as research advances and as we bring more tools forward to these builders.

[00:12:53] Drew Bent: And I'll just add on, we, we love this open approach. I mean, it's in the name Learning Commons to make it sort of this common g- you know, public good. And so It may sound strange, but we really don't want this Learning Commons and, like, grounded in the curriculum and the high-quality instructional materials to be a differentiator for Claude for Teachers.

Maybe it is today, but the hope is that the whole field, like no matter which edtech tool you're using or foundation AI model, that if you're using it for education use cases, that you have the context. It shouldn't be hallucinating the standards. It shouldn't be just vibe making the instructional materials.

It should be pulling from high quality open educational resources. And so we're very excited to see essentially what could be a race to the top where, you know, everyone could be leveraging, you know, when it makes sense, the Learning Commons data set. 

[00:13:44] Alex Sarlin: 100%, and I think that's been sort of part of Learning Commons' DNA, uh, from the outset is to create data sets and tool sets and, and ways for the entire field to, to raise its, uh, its quality bar when it comes to pedagogy and interoperability and sort of understanding how it's all connected all-- and how, what the underlying standards are as well as what the underlying learning components are.

It's relevant to many different people. One aspect of this project that I think we could extrapolate out on, because it's a really interesting one, is that there's a repository of agent skills for teaching that was released on GitHub by, uh, Anthropic, along with a technical write-up of how these teaching skills were evaluated.

So Drew, tell us about these skills that, you know, what, what are these teaching skills, and how is it different than how folks have used Claude to date? What, what, what do you think that these skills will do to enhance the pedagogical value of Claude for Teachers and, and maybe to your point about shared infrastructure, to increase the value of AI agents, uh, AI-enhanced education for everyone?

[00:14:41] Drew Bent: So th- these were co-built between Anthropic and Learning Commons. We both have had published them, and be- maybe taking a step back here, if you're thinking about how can we improve AI, you know, tools and models for teachers, part of it is bringing in the context, and so that's sort of the data layer that Sandra talked about with the standards, the learning components, the high-quality instructional materials.

There's also, of course, the context that a teacher can choose to put in from their class via, you know, Claude or Claude Cowork. So that's on the, the data side. But there's another layer here, which is you also sort of have to coach these AI tools on how to effectively, you know, differentiate instruction, make lesson plans, help an educator, you know, create a spreadsheet for small group instruction.

And so- As we found with these AI tools, like a lot comes out of the box, but it's also really helpful to build these essentially prompts or skills on top of it where you can grow the, grow the model's capabilities, and ideally have experts help design these skills. And so that's what we sometimes call, you know, agent skills.

And that's where we worked very closely with the Learning Commons team, bringing in experts from, from both organizations to iterate on these skills, you know, ground them in the learning science and the evidence of what works well for teacher workflows so that every teacher, when they're using Claude for Teachers or anyone else who chooses to adopt these skills, can leverage, you know, these insights.

And I'll say a big part and a big challenge of building skills is how do you know if it's making the, you know, the model better or worse? And so that's where you need to build in what we sometimes call evals or evaluations, essentially rubrics, where we are saying this is what makes really good instructional output.

And we publish all of that along with Learning Commons, both the skills but also those rubrics. And the hope there, you know, why do we publish all of this? The hope is that others can sort of take a look and say, "Okay, this is how Anthropic and Learning Commons have defined good pedagogy, and this is where I agree, and this is where maybe I disagree.

And I, you know, it's on GitHub, you know, I can make a pull request and suggest a change." So that's part of it, is to like sort of work out, you know, work in the public and be open about it so that we can get feedback on it. And then the other part is similar to the connectors and the data layer and the, the MCPs, which also can be available to the whole ecosystem.

The skills are also, you know, they're essentially markdown files that other edtech tools and other AI tools can adopt as well, and we hope to grow them over time. 

[00:17:18] Alex Sarlin: Yeah, that's really powerful. S-Sandra, what would you add in terms of how these skills were developed and what you hope that they will achieve in the ecosystem?

[00:17:26] Sandra Liu Huang: Maybe I can jump in a little bit to one of the specific skills that we worked on together. That's lesson differentiation. And if you work in education, that's like the thing that we all wanna do better. Educators are doing it every day. They're supporting a classroom full of students that are sometimes at different levels.

And, um, I'm really proud of the collective work that we did together because it is such a important need. And often, you know, we can-- especially AI tools by default might treat differentiation sort of as a approximate thing. Like, okay, that thing was last year, this thing is next year. Here's approximately a differentiated lesson plan.

But in the skill that we developed, you know, the experts across the organizations really were thoughtful about how do you actually leverage the learning progressions that we just t- we talked about earlier within the knowledge graph, knowing what skills come before and after. How do you actually differentiate a lesson even if not every student is actually achieving that standard quite yet?

And what we essentially are able to do is we don't simplify the task or the expectation, so the standard and the full expectations, those are-- it's really important that they stay intact, otherwise students don't actually get there. But what the, the lesson differentiation skill offers is essentially pulling the right supports that are needed, as well as the right, uh, acceleration that might be A student might be ready for to make sure that every level there's productive struggle that's still there, right?

That we're not taking away anything or any of the expectations for students, and that we're not stigmatizing any student if you're not quite there. But we're saying, "Hey, our expectation is really rigorous. This is the standard, but here are different levels of supports to help you get there or to help push you to the next level as well."

So, um, anyway, that's just a concrete example of, uh, pedagogy being captured in a skill being done openly, as Drew talked about, so that, you know, the field can continue to build from that and evolve it. 

[00:19:18] Alex Sarlin: That's ve-very powerful, uh, example and a powerful description about how this works. It's a, you know-- differentiation is something that people, you know, w-we use the word a lot, and there's lots of different proxies for that word.

But when you actually talk to, to expert educators about what differentiation looks like, they say exactly that. You cannot lower the standard or else the students who are not-- are behind will stay behind. They'll be doing easier work at a lower standard, and they'll never catch up to grade level. So the idea of capturing that, codifying it in a markdown file, which is in a skill which can be called by any educator, means that there's the ability to just differentiate at-- with that expertise built in.

That's incredibly exciting, and I think it's, it's a testament to the power of, of a system like Claude that has all of these skills baked in to be able to elevate pedagogy and, and sort of systematize it in a really interesting way. And then to both of your points, it's not fixed, right? I mean, differ-- that's the definition of differentiation as is.

A school or a teacher or a, a system might say, "Well, actually, we think differentiation is this." They might take the skill down off GitHub, fork it, say, uh, "This is our definition of differentiation," and then change the behavior down the line. It's really exciting to hear. So looking ahead a year or two, this is, I think, the beginning of a really interesting movement in education.

And what does success look like for Claude for Teachers? Um, what might the next phase of the work bring, whether that's is it gonna be a district offering deeper curriculum coverage, you know, or something you even haven't announced yet? You don't have to announce it here, but I'm curious how you're thinking about how this project will develop.

I think it's made a big splash already and sort of changed the, the game a little bit. How do you see it going next? Uh, Sandra, let me start with you on that. 

[00:20:54] Sandra Liu Huang: I think Claude for Teachers, we've been really excited, uh, especially for the collaboration with Drew and, and the team over at Anthropic, because I think not only are they trying to create great experiences for teachers, but really they're thinking about the whole ecosystem much in the same ways that we are at Learning Commons.

And so that just makes for a very exciting collaboration and one that I think can help the whole education field continue to, again, that race to the top, right? Where we can not spend our time navigating standards. Everyone's has to do that, but that, that can be something that's done once, it's maintained, it's high quality, it's open, it's transparent, and then everything that can hang off of that.

You know, that's something that we at Learning Commons will continue to invest in. I was talking to the team, you know, we have a million pieces of data in our knowledge graph, and a million is, is maybe small in the, in the LLM kind of territory, but that's a million pieces of, of expert-curated, high quality, rigorous connections that capture meaning and pedagogy that we will maintain for the field.

So we'll continue to build that out. We've got science and English language arts learning components and progressions coming to novel data sets around durable skills, which is a, you know, interesting emerging area that we need to, as a society, really grapple with and think about. And then we'll continue to partner, and Drew could talk more about Claude for Teachers, but thinking about how do we, um, deepen a- agent skills that we can build and offer, uh, for the field, and continue to really spur innovation and also just help educators have the really the powerful tooling that helps them be able to focus on things that just they have to do in the classroom with the students in front of them.

Um, hopefully we can really be a great partner to educators through these tools. 

[00:22:36] Alex Sarlin: That's exciting to hear. And, and Drew, h- how about you? What does success look like for Claude for Teachers, and what do you see next? 

[00:22:42] Drew Bent: Well, first of all, this is just the beginning. You know, we just recently launched this, and throughout the process of building it, we co-designed it with teachers.

We got educator feedback throughout, but again, still just the beginning. It's the summer right now. As we head into back to school, we're very excited to continue i- iterating with educators and experts on every part of it. So everything from the skills, like Sandra mentioned, adding on new MCP connectors, the deeper curriculum coverage, all of it.

Um, we're also working on a district offering, and we'll have more to share soon. We got a lot of feedback how important this is, and so we're excited to share more on that front. And then we think it's really important to go deep with a few districts, particularly in Title I settings, because we built Claude for Teachers to help raise the floor, particularly in some of the most resource-limited environments where teachers are doing a million things and they don't have a lot of time to always differentiate instruction they, like they would like to.

And so being able to help provide tools that can empower them, but then doing research to validate that this is actually helping them and figuring out the context in which, you know, a deeper partnership. We're working with Detroit Public Schools doing research studies to see how it's supporting teachers and ultimately helping them reallocate time back to student-facing work, which was a key thing as we've t- uh, discussed this with Detroit.

So we're very excited, just the beginning, and yeah, we appreciate you taking the time, Alex, to let us discuss this with you. 

[00:24:08] Alex Sarlin: Absolutely, my pleasure. I can't tell you how much the Edtech Insiders WhatsApp group and my LinkedIn channel and I think the entire edtech community has been really lit aflame by, by the potential of this, of this project and, and what it's doing to bring, uh, really sophisticated and very thoughtful AI-enhanced education into the hands of teachers, uh, all over the country.

It's really, really amazing. I just got back from Detroit last week. I'm incredibly excited to see what comes out of that, that sort of pilot-ish or, you know, the sort of deeper, deeper, uh, engagement that you're doing with Detroit to see what happens and w- how teachers react to it and what it does in, in classroom settings.

I appreciate you both being with me here, and I know that, you know, everybody is really excited about this project. It's incredibly powerful one for, for the entire edtech ecosystem. And I personally just love how collaborative it is. I mean, I think both of you are so well-networked and not, not only individually, but your companies work so consistently and deeply with so many different other people in the edtech space.

This really feels like even though it's, it's Claude for Teachers, it's really a, a sort of go-to-market and a, a movement that encapsulates a lot of different thinking from a lot of different orgs, from, from nonprofits to, to early-stage edtech AI native companies like Snorkel. It's really, really amazing. So thank you for being here.

Drew Bent leads education as part of Anthropic's Beneficial Deployments, and Sandra Liu Huang is president of Learning Commons, which funds and builds public AI datasets and resources to help bring more learning science into classrooms. Thank you both for being here with me on Edtech Insiders. Thanks for listening to this episode of Edtech Insiders.

If you like the podcast, remember to rate it and share it with others in the edtech community. For those who want even more Edtech Insider, subscribe to the free Edtech Insiders newsletter on Substack.

This season of Edtech Insiders is brought to you by Starbridge. Every year, K-12 districts and higher ed institutions spend over half a trillion dollars, but most sales teams miss the signals. Starbridge tracks early signs like board minutes, budget drafts, and strategic plans, and then helps you turn them into personalized outreach fast.

Win the deal before it hits the RFP stage. That's how top edtech teams stay ahead