'Built For Our Frameworks': How A Public School District Replaced EdTech With A DIY Platform And Coding Agents

Coding agents have made custom AI development cheap enough that public school districts can now build their own platforms rather than buying them. For schools with highly specific instructional philosophies, that matters: off-the-shelf EdTech tools are built for mass audiences and have always required compromise. In a K-12 district outside of Tacoma, Washington, that gap pushed the IT team to build its own platform from scratch, standing up lesson-planning, observation, data-querying, and coaching tools aligned to the district's own instructional frameworks. The governance question that follows is one every district will eventually face: What happens when anyone in the organization can build, not just IT?

Kris Hagel, CIO at Peninsula School District, has spent nearly two decades at the same district, moving from overhauling aging infrastructure to building the district's own AI platform from scratch. A Certified Education Technology Leader and CoSN board member, he brings both the technical depth of a working developer and the instructional grounding of a leader who has built curriculum-aligned systems from the ground up. That combination of practitioner-level building instinct and clear pedagogical conviction is what drove the district away from off-the-shelf tools entirely.

"We can customize everything to what we believe good instruction looks like, based on our own frameworks rather than what appeals to a mass audience," said Hagel. Peninsula's teaching and learning department maintains an instructional essentials guide grounded in the Danielson framework and Universal Design for Learning, specific enough that no commercial platform built to appeal broadly could reflect it accurately. That gap between what vendors offer, and what the district actually believes about good teaching is what set Peninsula on the path to building its own.

The district's in-house platform traces back to an open-source project from Vanderbilt University designed to let institutions safely interact with frontier AI models. When it became clear the platform was built for university researchers rather than K-12 classrooms, Hagel's team began building beyond it, using coding agents to develop tools aligned to the district's own instructional frameworks. What started as a workaround is now a full platform, tools for teachers, administrators, and district staff, each built around Peninsula's own instructional and operational standards.

• Pennies for prompts: The economics of that shift have changed faster than most districts realize. "Essentially, the code is cheap these days," Hagel said. "A $100 a month Claude Code subscription gets you a whole lot you can add to platforms." At that price, a six-to-eight-week build cycle is viable. The district's instructional coaching app, which lets teachers upload audio or video of their class and receive feedback against multiple pedagogical frameworks, went from concept to a pilot in about three weeks. "We put it in the self-service portal for any teacher to install on their laptop if they want to," Hagel said. "It's now available for all 600 teachers."

• Observation, automated: Peninsula couldn't afford instructional coaches in every building, so the platform filled a gap there too. "We built a tool just for them in our platform to do observations, help take some things off their plate as far as coding those and entering them into the evaluation tool," Hagel said. The tool also surfaces suggestions on how principals could improve their technique, turning a box-checking exercise into something more useful. When the superintendent wanted to add live audio conversations to the platform, the feature took an afternoon. "That functionality to have live audio conversations with AI and do that in a safe, secure manner, that was probably four hours to add," Hagel said.

Staff-driven experimentation is accelerating across the district, and not always within IT's line of sight. If IT doesn't provide safe pathways, staff will build anyway, just outside anyone's control, Hagel noted. Most districts are still working out how to respond, and Peninsula's answer is rooted in the infrastructure discipline the district has built over more than a decade.

• Front desk developers: Hagel described an executive assistant who built her own tool and quickly expanded it to require principal access. "It was one thing when it was your hobby project that just took something off your plate," he said. "But now that you've expanded it into something that can make a difference across the district, I've got to figure out how to deploy this safely and securely."

• Securing the sandbox: To manage that risk, Peninsula draws on a decade of running applications on AWS, applying cloud AI governance practices the team already knew. Student and demographic data flows through the platform with role-based permissions, each user seeing only what their role authorizes. "We have data processing addendums with the other providers that they don't train on our data, they don't store our data, any of that," Hagel noted. "When it comes to the LLM stuff, we're in pretty good shape." That security foundation opened up something unexpected: natural language access to the district's own student data. "My student database administrator now uses it to clean data," he said.

Building with coding agents is not uniformly fast. "I spent eight hours yesterday on a project that got nowhere, but I also spent 45 seconds on a project that would have probably taken six hours," Hagel noted. That variability is the backdrop for the district's next challenge: as enterprise tools for governing agent sprawl begin to emerge, Hagel is working through what it means to deploy autonomous agents on behalf of staff and what new infrastructure that requires. His answer is to build that infrastructure first and put it in front of people, rather than wait for use cases to justify the investment.

• Avatars and accounts: "What does it look like when we have to start creating email accounts and calendar accounts and GitHub accounts for all these agents that act on our behalf?" Hagel asked. "There's a whole other set of infrastructure that school districts have to start thinking about now." He has spent weeks building out OpenCLAU agents inside segmented VMs on the district's AWS infrastructure, tightly controlling what they can access. There is no established playbook for this at the K-12 level, and Hagel is writing it as he goes.

• Infrastructure first: His deployment strategy reflects that same philosophy. Next week, the executive cabinet gets access to agents for the first time. In May, early-adopter teachers follow. The goal is not to define what agents are for. It is to put them in front of people and let the use cases emerge. "I was having a conversation with some CEOs of some EdTech companies last week, and they said, 'Tell me what the use case is, and then I'll build that for you,'" Hagel noted. "I don't know what the use case is. I'm going to build the infrastructure for people," he added, "and then let them try it, then they will tell me what it works for."

That build-first philosophy is already reaching the district's back office. When the enrollment coordinator was preparing to leave, Hagel and the CFO mapped out how much of the job was rules-based and handed it to an agent. The district publishes everything it builds to its public GitHub, free for any school to use. "I sat in the vendor hall at the conference last week, and I looked around and said, I can rebuild that in three days," Hagel said. "Your product is way more complicated. I don't even know that I would try and tackle it," he continued, "but that doesn't mean somebody else isn't going to."

For Hagel, the question is no longer whether districts can build. It's what that means for the EdTech market going forward. "Some of the really forward-thinking EdTech companies are thinking about what they need to do to change," Hagel concluded. "It'll be interesting to see where the whole thing goes, what can districts do on their own, what do they need their partners for, and where do the partners adapt and adjust moving forward."