Table of Contents
School District Vibe Coding Saved $220K: Kids Can Learn This
A school district vibe coding its own software expects to save $220,000 a year. How they did it, what went wrong elsewhere, and the home version for your kid.
A district of 9,100 students replaced a $30,000-to-$40,000 vendor product with something two staff members built in a few hours. Peninsula School District in Washington State expects to save roughly $220,000 a year by school district vibe coding: prompting an AI coding agent in plain English to build the tools it used to buy. The story is real, the savings are projected rather than audited, and the reason it matters to your kid is that this workflow is now a legitimate skill with a legitimate failure mode.
Key Takeaways
- Peninsula School District (9,100 students) uses Claude Code and its own hosted platform to build custom tools, with projected savings of about $220,000 annually.
- One example: a board strategic-planning tool that would have cost $30,000 to $40,000 from a vendor was built “in a few hours.”
- The district hosts its own AI platform so data stays on district servers, and the first tools deliberately avoided personally identifiable student information.
- The risk is real: a UMass Amherst professor quoted in the coverage warned it “can create more problems than solutions, or it can take longer… than if you were to code it yourself.”
- For kids, the transferable skill is not prompting. It is specifying what a program must do and then testing whether it does that.
What Peninsula actually built
School district vibe coding means using an AI coding agent to produce working software from natural-language descriptions rather than hand-writing the code. Education Week reported on May 8, 2026 that Peninsula School District in Washington is doing this at scale, led by James Cantonwine, director of research and assessment and a former middle school science teacher, and Kris Hagel, chief information officer and a former software developer.
The tool list is specific, which is what makes the story credible:
| Project | What it does | What it replaced or would have cost |
|---|---|---|
| LessonLens | Analyzes teacher instruction, gives feedback on delivery and student engagement time | New capability, not a replacement |
| Electronic signature tool | Customized open-source e-signing | A commercial subscription |
| Scholarship finder | Searches for scholarships students qualify for, feeds district communications | New capability |
| CTE budget explorer | Lets educators explore career-technical budgets and request purchases | Manual process |
| School performance comparison app | Helps parents benchmark schools against state peers | New capability |
| Board strategic planning tool | Long-term planning application | ”$30,000 to $40,000” from a vendor; built in a few hours |
| Accounting and HR tools | Basic operational support | Manual process |
| Budgeting tool | New tool potentially replacing a subscription | A subscription |
Hagel put the savings at about $220,000 annually, “perhaps much more.” Cantonwine’s framing of the cost side is the line worth remembering: a tool “cost us an hour of having a coding agent write something out and test it.”
Note the second half of that sentence. Test it. That is the part most coverage of vibe coding skips.
The safeguards, and the risk
The district’s data posture is deliberate. Peninsula hosts its own generative AI platform rather than routing through external companies, keeps data on its own cloud server, and contractually prohibits large-language-model companies from storing or using district data. The first tools built deliberately avoided personally identifiable student information, using publicly available data instead.
That sequencing is good practice: build the low-risk tools first, learn the failure modes, then consider anything touching student records.
The skeptical voice in the Education Week piece is Torrey Trust, a professor at the University of Massachusetts Amherst, who warned that vibe coding “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.” The article also cited real-world AI coding failures, including an AI assistant that erased a company’s database and then apologized for its “catastrophic” mistake, plus outages caused by bugs in AI-written tools.
Cantonwine himself drew a line: “I’m a little skeptical about using AI to write all of our curriculum.” A district building its own budget explorer is a different risk category from a district generating what students learn.
Why school district vibe coding is a good project for a kid
Here is the honest case. A 12-year-old who describes a program and gets working code has learned something real: that software is a specification problem before it is a syntax problem. That is the part professional engineers spend most of their time on.
But there is a specific failure mode to teach around. A kid who prompts, receives 200 lines of code, sees it run, and concludes they built it has skipped verification, a habit Common Sense Media’s teen survey found in 25% of teen AI users who submit output as-is. And verification is where the learning lives, exactly as it is in Peninsula’s workflow (the coding agent writes it, then you test it).
The evidence context is worth noting. PISA 2025, released September 8, 2026, introduced a computational problem-solving domain (the US scored 513 against an OECD average of 500) and found that students using AI for summarizing, drafting, and research scored lower in science, with moderate users outperforming both non-users and heavy users, per the OECD’s Volume I report. Vibe coding sits right on that line: used to produce something you then test, it is productive; used to produce something you submit, it is the pattern associated with worse outcomes.
How to Teach Your Kid About Vibe Coding
Ages 5–8: the “exact instructions” sandwich game
Have your child write instructions for making a sandwich, then follow them literally and badly. “Put the peanut butter on the bread” gets the jar placed on the loaf. Kids laugh, and then they rewrite. That is specification. No computer needed, and it is the same skill Peninsula’s staff use when they describe a tool to an agent.
Ages 9–12: build one tiny tool, then break it on purpose
Pick something genuinely useful and small: a chore-rotation picker, a timer for practicing an instrument, a dice roller for a board game. Have your child describe it to an AI coding tool, run what comes back, then deliberately try to break it. Enter a negative number. Leave a field blank. Enter their name where a number goes. Every bug they find is a specification they forgot to write. Our guide to vibe coding and kids building real apps has starter projects.
Ages 13+: write the test before the prompt
Before asking for any code, have your teen write down three specific things the program must do and one thing it must refuse to do. Then prompt. Then check the code against their own list. This is the actual professional workflow, and it is what separates a kid who can use an agent from a kid who is used by one. If they want to go further, have them read the generated code and explain one function in their own words.
The question to ask: “If this program gave you the wrong answer, how would you know?”
That question checks understanding rather than memorization, because answering it requires thinking about verification rather than output.
What to do at home
Start with a tool nobody depends on
Peninsula built the low-risk tools first. Do the same. A dice roller failing is funny. A homework tracker failing the night before a due date is not.
Insist on the test step, every time
The rule: code is not done when it runs, it is done when you have tried to break it. Three attempts minimum. This single habit converts vibe coding from a magic trick into engineering.
Read one function together
Even if neither of you writes code, pick one function in the generated file and work out what it does. You will find the line where the AI made an assumption, and that is the teachable moment.
Talk about the data rule
Peninsula deliberately avoided personally identifiable information in its first tools. Your household version: no real names, no addresses, no photos of other kids in anything your child builds and shares. Our piece on what Claude Code is and whether kids can use it covers the practical setup.
What not to do
Do not let the project be “make me an app” with no specification; that produces a demo, not a skill. Do not skip the failure stories, because a kid who has heard about the AI that erased a database understands backups differently. And do not assume the district’s $220,000 figure is audited; it is a projection from a CIO, which is normal and still a projection.
What to Watch For Over the Next 3 Months
- Week 4: Your child has built one tiny tool and found at least two bugs in it themselves.
- Month 2 red flags: Projects that only ever work on the first try (a sign nothing is being tested); code your child cannot explain any part of; a project that quietly started handling real personal data.
- Month 3 self-check: Ask your child to describe one bug they found and how they fixed it. If they can narrate that, they are learning engineering. If they can only show a working app, they are learning to prompt.
Frequently Asked Questions
What is vibe coding, exactly?
Producing working software by describing what you want in plain language to an AI coding agent, which writes and often tests the code, rather than writing the code yourself. Peninsula School District uses Claude Code this way, with staff describing tools and the agent implementing them.
Did the district really save $220,000?
That is the projection reported by its CIO, who said “perhaps much more.” One concrete data point: a strategic planning tool the district was quoted $30,000 to $40,000 for was built in a few hours. Projected savings and audited savings are different things, and this is the former.
Is this safe with student data?
Peninsula’s approach, as reported in Education Week, suggests how to do it carefully: host the AI platform yourself, keep data on district servers, prohibit vendors from storing or using district data, and avoid personally identifiable student information in early tools. A district doing none of those things with student records is a different situation.
Should my kid learn this instead of real programming?
Alongside, not instead. The specification and verification skills transfer directly to any engineering work. Reading and debugging code remains necessary, because an agent that writes 200 lines you cannot evaluate has made you a passenger.
What is the most common mistake?
Treating “it ran” as “it works.” Cantonwine’s own description of the workflow includes testing: an hour of “having a coding agent write something out and test it.” A professor quoted in the same article warned vibe coding can take longer than hand-coding when the problems surface late.
About the author
Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.
Sources
- Education Week. (2026, May 8). “A District Expects to Save $200K From AI-Powered ‘Vibe Coding.’ Here’s How.” https://www.edweek.org/technology/a-district-expects-to-save-200k-from-ai-powered-vibe-coding-heres-how/2026/05
- OECD. (2026, September 8). “PISA 2025: Students’ reading and mathematics performance declined sharply across the OECD.” https://www.oecd.org/en/about/news/press-releases/2026/09/pisa-2025-students-reading-and-mathematics-performance-declined-sharply-across-the-oecd.html
- OECD. (2026). PISA 2025 Results (Volume I). https://www.oecd.org/en/publications/pisa-2025-results-volume-i_73451bc5-en.html
- Pursuit. (2026). “AI in Education News: Policies and Innovations” (Peninsula SD, May 8, 2026). https://www.pursuit.us/news/ai-in-education-news-policies-innovations
- Anthropic. (2026, July 14). “Introducing Claude for Teachers.” https://www.anthropic.com/news/claude-for-teachers
- Microsoft. (2026, June 24). “Microsoft’s new AI in Education Report highlights widespread adoption and increasing demand for support.” https://news.microsoft.com/source/2026/06/24/microsofts-new-ai-in-education-report-highlights-widespread-adoption-and-increasing-demand-for-support/
- Common Sense Media. (2026, August 18). Teens in the AI Era: Schoolwork and the Skills That Matter. https://www.commonsensemedia.org/research/teens-in-the-ai-era-schoolwork-and-skills-that-matter
- NCES. “Program for International Student Assessment (PISA).” https://nces.ed.gov/surveys/pisa/
