How many times have you had the best idea but you didn’t know where to start? You tell yourself that you aren’t “good” at business/law/technology, so you let the cost and risk of starting– wasting time and money, looking stupid, not knowing where to begin– wither your concept on the vine. But what if there were an easy way to make your vision tangible to others? What if instead of weeks or months of toil, you could prototype at the speed of thought? It turns out there is a way. Part pejorative, part magical, vibe coding is the use of Large Language Models (Claude is my favorite) to build software without knowing how to code. I chat with it all day—asking questions, getting code snippets, pasting them line by line into what’s now Schutzduck, the AI advisor I created. My wife asks me if I talk to a robot all day with a concerned look in her eyes. Kind of, yeah. Personally, I’ve always struggled with working alone. I get a burst of energy to create something: a Schutzblog draft, a business idea, even where to go for dinner, but the flow always dries up, leaving me stuck. I hit the limit of what I can do alone, knowing that more effort won’t help, but the work still isn’t done. Rather than give up, I’ve learned this is my signal to search for feedback: coworkers, customers, poor unsuspecting patients, my wife and children. Each conversation helps me iterate, but after two or three rounds I notice the same look in everyone’s eyes when they get hit with the “Schutzbank firehose.” I’m ready for round seven, but my partners are ready to tap out. Don’t miss out on the latest Schutzblog, subscribe now! Paid subscribers get early access to new content + the full archive. Maybe you’ve felt this too. But when you build with Claude, it lets you free your ideas from the prison of your mind. You share an outline of your idea, and it starts to add the missing pieces–the JavaScript you don’t know, the pictures you can’t draw, the regulations you are terrified to break. Each round of question and answer refines your idea, crystallizing it into a form that you’ll know is right when you see it, but you could not have created alone. You are able to give feedback on your own work. As a benefit, because the AI did the work, you don’t fall in love with bad versions. No sunk cost, no endowment effect, no guilt from demanding 100 wedding cakes when no baker is suffering—just sustained creative flow that turns concepts into prototypes. And it turns out that concepts die unless they become prototypes. The distance from concept to working product used to require trading your equity (at a steep discount) for outside funding, so that you can hire (and manage, ugh) teams of expensive people all before you even know if the thing works or if anyone wants it. Yuck. But now you can quickly and cheaply build a prototype by yourself, unencumbered. That’s the superpower, the promise of vibe coding: not drone armies replacing us, but power armor—amplifying what each of us can create in ways that were too expensive, too slow, or too dilutive to attempt before. So what could you imagine doing? What have you been putting off because, “reasons”? Let’s not pretend that vibe coding is unmitigated upside. LLMs hallucinate, they confidently spit out wrong answers, they zone out when you flood them with information. Who cares? You know who else does that? People. “It works on my machine.” “I already checked that!” All things you’ve heard from your expert teammates, and you shrug and roll with it. Understanding the veracity of information is still your job as a creator. Clinicians already know this lesson well—we repeat labs when we don’t like the results, compulsively recheck medication lists, and ask historical questions five different ways when things don’t add up. Apply this discipline to your vibe coding. If it doesn’t make sense, ask again. Or ask a different LLM. When the conversation starts to wobble, start a new one. Unlike people, it won’t get angry, impatient, or defensive. It can’t. So stop worrying how far these tools can go and start asking how easily you can begin. So what have I done with my new suit of power armor? * Calendar sync: built a Google Apps Script to allow my clients to see my availability without seeing my business. Just set my calendar and forget it. Blows Calendly out of the water. * Winter tire sensor: I finally grew up and got winter tires, but when to put them on? When the average temperature is under 45°F for 5 out of 7 days in a row. Who has time to keep track of that? Yup, another Google Apps Script that emails me when it is time to change them. * Schutzbot: Schutzblog is a knowledge library, but it is published like a periodical. Schutzbot serves as the librarian, directing you to the most relevant articles, rather than the most recent. I’m pretty sure it is the only chatbot hacked onto Substack. Maybe next time I’ll tell you how. * Schutzduck: an AI advisor trained on my writing—wise, irreverent, available 24/7 to help you get unstuck on that big decision. I built the whole SaaS business (auth, payments, usage tracking) iteratively with Claude. Feeling stuck? Chat with the duck. This is a rallying cry for the creative-yet-technically-limited. Those who dream in colors but cannot draw. But now you can! Make your mad visions real enough that you—or someone who knows what they’re doing—can take them further. Skip the awkward “what does that word mean?” part of learning a new skill, because you can teach yourself by doing. Your goal isn’t perfection—it’s either “works good enough for me” or “clear enough to hand off.” Both are wins. Don’t know where to start? Pick one problem. Open Claude. Describe it. See what happens! Don’t miss out on the latest Schutzblog, subscribe now! Paid subscribers get early access to new content + the full archive. Get full access to Schutzblog at www.schutzblog.com/subscribe
