I had gone to sleep feeling excited because I’d finally launched my software to help parents in Sweden navigate school and medical services. In three months, I used Claude Code to write the documents, help me create the code, and launch a social media campaign (while simultaneously managing our family’s school and medical contacts). As someone who had never coded before, I was really proud of what I created.
The next morning, I checked my phone and saw fourteen receipt emails from Anthropic, the company behind Claude. I knew something had gone terribly wrong. While I was asleep, a bot had found a way into my app’s backend and gone on a $686 shopping spree using a Claude API key I didn’t even know was still sitting exposed on my first version of the software.
The first thing I did was open my regular Claude agent, the one that had helped me build the project. Together, we searched the security logs and found the source of the attack. So there I was, using one AI to talk to another AI. My regular Claude agent prepared all the details of the attack to pass off to the anthropic billing agent.
What came next felt like the Swedish medical circle of “we’d love to help you, but you have to go over there.” And then you go over there, and they tell you to “go back to the first place”, and when none of it works, you ask someone else to verify who you’re even supposed to be talking to, and the third person doesn’t even know. So there was a perfect irony in building software meant to help people navigate a circular system, only to get stuck in the same kind of loop with the AI agents.
Can I Talk To A HUMAN?
Here’s roughly how it went. I explained the situation clearly, with evidence. An attacker’s IP. A statistically impossible spike in usage. The bot was warm and sympathetic. It told me my case “clearly warranted” human review and that a human agent could look into this. It even said it was “transitioning” me to a human support agent.
So I waited. And chatted with my other AI assistant, collecting more evidence of the fraud in the meantime.
I asked again. Same “we care about you” tone and a vague reference toward a human who might, possibly, look at my case.
When I knew it was going in circles, I asked directly: “Has this actually been sent to a person, or am I still talking to a bot?”
The bot told me it was an AI assistant and could not, in fact, escalate to a human. But it did reassure me that talking to a human would be a good idea.
I tried several variations of wording to try to get a human agent on the line. At one point, I was told to escalate through a different channel. That channel told me to go back to the first one. Several messages, across two separate threads, over a few days. Each one acknowledged warmly. Three weeks later, I still haven’t gotten a human reply.
The System Loop
It seems all the systems are failing, and we are stuck in the middle.
Right before this AI debacle, I was juggling a denied claim with Sweden’s social insurance agency (which we did get overturned only after filing a ton of paperwork) and an appeal related to denied accommodations for school. At the same time, in the medical system, the specialist referrals kept getting returned, and I had a doctor asking me what to do because the system wouldn’t help my child get a blood test. Did I mention our house is also for sale?
So when the AI support bot kept warmly redirecting me in circles, it wasn’t an isolated tech failure. It was one more loop, layered on top of all the other loops I was already standing in.
The whole system felt like it was collapsing around me while the chatbot was busy trying to be “helpful.” After I thought I’d finally found the correct thread to get in line for a human, I stopped replying to the emails. The support bot did a check-in: just wondering if you were doing ok?
I’m thinking no. I’m not doing ok. But your extra check-in isn’t moving the needle at all here. Cue me laughing.
Maybe Things Will Work Out
That day, I felt the anxiety and general dread of nothing working and too many people to contact. I knew I’d find a way through eventually, because that’s just what I do. But I also felt genuinely defeated.
I was the founder of an AI navigation tool meant to help overwhelmed parents cut through bureaucratic nonsense, talking to an AI chatbot to help manage my other AI chatbots, while fighting three other bureaucracies at the same time. It took a while before I could zoom out and see the humor in it all.
I had to sit with it for a bit. Make the reports that had deadlines attached. And rest before making my next move.
What I Actually Learned (Besides “Lock Down Your API Keys”)
I started this whole project because I was tired of navigating things alone. I want to build a community around real solutions for the medical and school problems I keep running into. I want to use strategies that are actually working to help support other parents. The plan was simple: sign up, chat with an AI agent, and get navigation advice.
I don’t regret building any of it. It was a genuine learning experience. The way I have gotten my biggest breakthroughs in our own family struggle is to learn to ask AI sharper, more specific questions myself. While researching and compiling the documents that became the backbone of this project, I learned a lot that directly helped our case. Something that I know from being a teacher for so many years before becoming a parent: the best way to become really good at something is to teach it to someone else.
So here’s the pivot
Instead of building another product with an API key that can get stolen, I’m packaging up the actual method, the specific questions, and the way of using AI that took me a long time to figure out, so others can use it with whatever AI they trust. Claude today, something else next year, doesn’t matter. You’re not locked into my app. You will own the lessons learned and will be free to apply them in other areas of your life.
And for those who don’t want anything to do with AI at all, I’m also working on a book, so you can write your own documents, walk into your own meetings, and build the confidence to think through your case yourself, instead of staying stuck without school or medical care. We know nothing will work every time. But taking steps in the right direction is still how things move forward.
Turns out the most useful thing I built in the past few years wasn’t an app. It was a hard-won understanding of the system here and what resources we are entitled to. A confidence that comes from understanding your rights and not being afraid to ask for what you need. That has helped me to keep stretching and putting together a team around my family. Part of that team has been an AI assistant. Part of it has been the humans actually working to make our family heard and seen.
Stick around. We’ll be launching the Skolnavi system soon, with prompts and guidance built from everything I just described. It’ll be worth the wait. I promise. No bad music while you sit on hold.