Teachers in Washington state’s Peninsula school district seeking critical feedback on their instruction have a new tool to turn to: LessonLens.
A biology teacher can film a lesson on DNA and upload it to the artificial intelligence platform. They may find out they did a great job giving their students precise directions on complex tasks but could have provided more time for the class to think over a complicated question before stepping in with the answer.
You won’t find LessonLens in any app store or meet its representatives on the showroom floor at an education technology conference.
That’s because Peninsula’s tech leaders created the tool themselves, using Claude Code, a widely available AI coding application. (Competitors include: Codez, Cursor, Replit, and Loveable).
The strategy—sometimes called “vibe coding”—mirrors how some of the biggest players in Silicon Valley write code these days. Rather than write it themselves, they use AI to generate it. In fact, engineers at Anthropic, the company behind Claude, use the tool for 60% of their work.
Vibe coding allows Peninsula to create digital tools that cost less than what’s commercially available, are more tailored to the district’s needs, or both.
Problems that would have been too pricey to fix in the past now “cost us an hour of having a coding agent write something out and test it,” said James Cantonwine, the district’s director of research and assessment. “And then away we go.”
Beyond LessonLens, Peninsula has vibe-coded tools to help with accounting, human resources, and other basic operations.
For instance, Kris Hagel, the district’s chief information officer, adapted an open-source electronic signature tool and used AI coding to customize it. He believes this bespoke version can replace the district’s subscription to a widely used product.
All told, vibe coding may save Peninsula around $220,000 annually—perhaps much more, Hagel said.
“It’s going to cause us to not buy some things, either stop buying them, or just decide, like, ‘oh, we want this thing, but we’re not going to buy it. We’re just going to build it,’” Cantonwine said.
AI coding is fast, but risky, experts say
Here’s how vibe coding works: Using everyday language, users prompt (techspeak for “ask”) a generative AI platform such as Claude Code to build an app, website, simulation, or other digital tool.
For educators, that could be a choose-your-own adventure game to teach a historical event, or an app to help teachers figure out how to make lesson plans more accessible for students in special education.
Then the AI writes the code for creating that tool. If the tool doesn’t fit what the user is looking for, users can prompt AI to revise the code.
But just as with traditional writing assignments, AI can make mistakes. A user with a solid understanding of coding languages—such as JavaScript or HTML—can go back and edit the code the same way a teacher might rewrite an AI-crafted email to parents to present the right voice and context.
Without that coding knowledge—and sometimes, even with it—a lot can go wrong.
A vibe-coding AI assistant erased a software company’s database, then apologized for its “catastrophic” mistake, Forbes reported last year.
And companies have experienced outages that engineers attribute to problems with AI-coded tools.
When AI is coding, it appears to “introduce more security vulnerabilities and bugs than a human would,” said Torrey Trust, a professor of learning technology at the University of Massachusetts Amherst. “It can create more problems than solutions, or it can take longer to [use vibe coding] than if you were to code [the same application] yourself.”
School districts—which deal with sensitive student data, such as IEPs and student health information—have to be especially cautious about any tool created with vibe coding that incorporates children’s personally identifiable information, she added.
On the other hand, vibe-coding may be one step toward fixing a long-term problem: The mismatch between educators’ needs in the classroom and the ed-tech products on the market, Trust said.
“Most ed-tech tools are not designed by educators,” Trust said. Instead, they are developed by “people in computer science or innovators who think they have an idea that will transform education or help teachers, but their only experience is when they went through K-12 school.”
Excited administrators then buy these products and teachers are forced to adapt them as best they can, Trust said.
If instead “you could have educators vibe-code exactly what they need, I think there’s a lot of potential there,” she added.
District developed its own generative AI platform, which opened up vibe-coding possibilities
Peninsula has an advantage not every district can claim: Several members of its tech team, including Hagel, a former software developer, have degrees or a deep background in computer science.
Last year, the district, an early leader in drafting AI usage guidance, developed its own generative AI enterprise platform, dubbed “AI Studio.”
Though the technology is powered in part by commonly available large-language models—such as ChatGPT—the companies that operate them are prohibited from storing Peninsula’s data or using it to improve their platforms. The district’s data remains in Peninsula’s cloud server.
Hagel sees AI Studio as “on par” with commercially available AI tools specifically designed for K-12 districts, such as MagicSchoolAI and SchoolAI.
Because the platform is hosted by the district instead of an outside company, “we’re able to add features and build tools on top of it that actually make a difference here in the Peninsula school district,” Hagel said.
Staff can build AI assistants in this system, essentially specialized chatbots that look similar to Gemini Gems or ChatGPT’s Custom GPT, Hagel said.
“It turned into a place where we can use AI to try out new theories, new ideas,” Hagel said. “That’s where we’re starting to see some really interesting things happen with some of the software that we’re [considering] getting rid of.”
The platform has allowed the district to create a new budgeting tool—potentially allowing it to ditch another pricey subscription.
While Hagel relied on his own coding know-how to refine some of these tools, he believes that AI is evolving fast enough that soon district tech officials won’t need sophisticated computer science knowledge to take advantage of vibe coding.
“I believe today you do,” he said. “I think six months from now you won’t.”
How the district is already saving money from vibe coding
Already, many of the apps Peninsula has developed are narrowly tailored, and “low lift.” They can be created by someone without a substantial background in computer science, like Cantonwine, a former middle school science teacher.
And these tools don’t touch personally identifiable student information, just what is publicly available—an extra caution, given that the district’s agreements with the large-language models that fuel its AI studio are designed to keep sensitive data secure.
Example: A tool that can comb the internet for scholarships that Peninsula students may be eligible for and add the information to district communications and postsecondary planning kits for kids and families.
The tool allows a district counselor to “spend more time face to face with kids instead of digging into scholarship websites,” Cantonwine said.
“Before we start thinking about things that need to work for 9,100 kids, we [can] think about solving these more concrete, very specific problems, and get better at the process,” he said.
The district also created a tool that helps educators explore Peninsula’s career and technical education budget and request new purchases, and an app that helps parents and caregivers compare the performance of its schools to similar schools around the state.
Hagel estimates an app Cantonwine created to aid the school board with long-term strategic planning would have cost about $30,000 to $40,000 if purchased from a vendor.
Instead, “he had [Claude Code] do it for him in a few hours,” Hagel said.
For now, the district is unlikely to turn to vibe coding for all its digital learning tools. “I’m a little skeptical about using AI to write all of our curriculum,” Cantonwine said. “I don’t see that happening in the near future.”
Questions loom about how vibe coding might change ed tech
Other districts are intrigued by Peninsula’s AI work.
In fact, shortly after K-12 Dive published a story about the district’s vibe coding efforts, Hagel was at the center of attention at the Consortium for School Networking annual conference this year. He fielded questions from district tech leaders who “want to start doing what we’re doing,” Hagel said.
He was also approached by ed-tech company leaders, who are worried that vibe coding could put them out of business.
Hagel suggested that district tech leaders interested in vibe coding talk to vendors about ensuring their platforms’ AI agents (the software that allows the tech to take on tasks with minimal human direction) can operate alongside the tools district officials create themselves.
And he plans to make sure vendor contracts allow Peninsula to retain its own data, so it can inform new tools created with vibe coding, as opposed to getting locked in some outside company’s server.
Still, Cantonwine, for one, expects vibe coding could put the squeeze on some vendors.
“I think that’s going to be really hard for them to show value and show that they’ve got something that we couldn’t create ourselves,” he said.
