The Door That Wasn't There

For the past few years, I have been trying to build things with AI. Custom ChatGPTs. Basic vibe coding experiments. Small automations that half-worked and taught me a lot anyway.

I always described vibe coding for non-techies as communicating with someone who speaks a different language using your phone as a translator. It works until the other person suddenly smacks you. And you never quite know where the translation went wrong.

Copy the error, paste it back, and ask for a fix. Then hope and pray. That was the loop. It worked, sometimes. But limited context windows, tokens, and patience made it frustrating.

Release after release, the tools got good enough and now catch their own mistakes, so you can see the error, understand what broke, and correct it without having to play translator. The loop got tighter, and the smacking stopped.

So, last week I built a “full” application.

For me, the door was never locked. It simply didn’t exist until one afternoon, it did.

What I built

If you want professional software to write a book, you’ve had two real options: fight Microsoft Word for three hundred pages, or pay for specialised writing software, spend time learning it, and use about a third of what it offers.

I used Scrivener for years. It’s genuinely good. But what I built last week is better…for me. It has everything Scrivener has: chapter navigation, progress tracking, word counts, and version snapshots. And things Scrivener has never had: an AI co-writing assistant trained on the book’s own framework, a Red Team mode that destroys my arguments, a research panel, an ideas scratchpad, and an automated backup that syncs across machines via my Google Drive.

It is exactly what I needed, minus features I’ll never use. No licence to manage.

Scrivener took a small team years to build. I built this in an afternoon — with decades of experience in marketing, sales, and HR behind me and zero training in software development.

How? And why me, specifically?

The pain was real. This wasn’t a hobby project. I write every day. The frustration I felt with existing tools was specific. When a problem genuinely matters to you, you don’t give up when the first version breaks. How much you iterate is a direct function of how much you care about what you’re building.

Know what you want. This is the condition most people underestimate. I didn’t need to be a developer. But I needed to know what a good writing tool looks like, and I did, because I’d been using them for 20 years. Vague prompts produce vague tools. Domain knowledge is the real leverage.

Structure the problem. The book has an index: twenty chapters in four parts, each with a brief. That structure became the backbone of the tool. The people who get the most from AI-assisted building are those who’ve already done the intellectual work of structuring their problem and their information. Vibe coding taught me the hard way.

The thing that matters most

The tool takes automated backups every five minutes, keeps daily archives, and lets me export everything — content, sources, notes, chat history — at any point. If Claude doubles my subscription fee and introduces ads tomorrow, I have everything I wrote in a format I can open in any text editor.

That’s not a technical detail. Your drafts, your research, your thinking (yes, AI augmented) — these are your IP. You built it; you keep it.

What this means

This is not just about writing apps.

A simple agent that scans the news and drops a summary into your inbox. A tool that tracks your client notes and flags when someone hasn’t heard from you. The range goes from a one-hour automation to a full afternoon application. The common thread remains: no product team knows your workflow as well as you do. And now that knowledge is “executable”.

The people who thrive in this shift are not the most technically literate. They are the most precise about their own problems. That comes from domain knowledge, from noticing what slows you down rather than just accepting it, and from the willingness to treat the first version as a beginning, not an end.

The category of things I once assumed were “not for me” just got smaller.

Stop learning and testing 1000+ AI tools or waiting for the right one to come out. You can now build solutions to your own problems.