Death by a Thousand Free Tiers
I run a moderately successful technical writing agency. Over five years, I had accumulated the kind of tooling stack that makes you wince every time you try to explain it to someone new.
- Trello — Project management for client work
- Linear — My personal CRM for sales and hiring
- Spreadsheets — Filling in the gaps for reporting and analytics
- Vibe-coded Flask App — Pulling data from Trello for yet more reporting
- GitHub Labels + Custom Scripts — Various automations held together by hope
It was an evolved collection of tooling, each piece solving a problem as I identified it. Organic growth. Technical debt made manifest. The kind of system that works perfectly fine until it doesn't, and when it doesn't, you can't remember which spreadsheet formula is connected to which Zapier webhook.
I wanted something better. Something that was mine.
Bridge: Built for One User
If you want to try it out, it's at bridge.ritza.co. It's a kanban app with some special features—specifically the things I care about most, which probably won't be what most other people care about:
- Blazingly Fast — Like Linear used to be before they added all the features. Instant response. No spinners.
- No Per-Seat Pricing — I hit my issue limit on Linear's free tier. Never again.
- Not Bloated — Unlike what Trello became. Just the features I need, nothing else.
- Real Markdown — Including image upload from clipboard. Because that's how I think.
- Card History — Full tracking for each card. Who changed what, when.
- Dynamic Rules — Populate fields from other boards. My 'Hiring' CRM feeds into assignees automatically.
How I Built It
The stack: Django SaaS Pegasus ($50), Claude Pro ($20/month), Codex ($20/month), Hetzner VPS ($5/month).
Claude Code, Codex, and Pi tools—they're all similar enough that it doesn't matter which one you start with. I hit my limit on one and switch to another. The experience is remarkably consistent.
The Dangerous Part
Here's where I lose some of you: I give Claude and Codex root access to my VPS. They can deploy each feature to production immediately, and I try it out there.
Yes, it's dangerous. No, it's not appropriate for enterprise software or financial systems or anything super important. But the risks are very acceptable to me—and I think to a lot of people who shy away from them because we've all been taught it's a bad thing to do.
There's actually a psychological principle that makes me question my confidence here: the Peltzman Effect, also known as risk compensation theory. The more often someone does something risky without negative consequences, the more likely they are to take excessive risks in the future. Named after economist Sam Peltzman, who found that drivers compensated for improved car safety features by driving more dangerously.
Maybe I'm just lucky so far. Maybe the disaster is coming. But for now, the velocity gains are real, and the worst case is I reinstall a $5/month VPS.
Why You Should Build Your Own Software
Not for everyone. Not for every use case. But for small-business or personal software, the calculus has fundamentally changed.
- It's easier than ever. The gap between "idea" and "working prototype" has collapsed. What took weeks now takes hours.
- You'll actually enjoy using it. There's something different about software you built yourself. Every quirk is intentional. Every feature exists because you needed it.
- The economics work. A domain name, a VPS, a Claude or Codex subscription. Add the other one when you hit your limit. That's it.
Get a domain. Get a VPS. Get a Claude or Codex subscription. Start building something you actually want to use every day. See Bridge in Action.
Skills That Still Matter in the Age of AI
There's an open question about what AI is becoming. The answer determines everything about what skills you should be developing.
Is AI a compiler? If so, technical skills become more valuable. Everyone with an idea can now build, so execution quality becomes the differentiator. Or is AI a printing press? If so, maybe no one will be hiring programmers in five years. The skill becomes table stakes, like literacy.
I don't know which it is. But either way, some skills seem more durable than others.
UX & Product Intuition
You need to make software delightful. That's about intuition, experience, empathy. AI is still bad at this stuff. It can generate a login form, but it can't feel the friction of an extra click or notice when a transition feels janky. The "product engineer" who cares about these details is more valuable than ever.
Testing & Verification
Agents are still remarkably bad at testing their own work. I've seen impressive setups that automate a lot, but they use a lot of tokens, take a long time, and are brittle. Maybe someone has solved this. But I'm still using my intuition to think "hmm, this could be broken, let me try it"—which also helps with the "delightful" part above.
Noticing the Rough Edges
Only by using software extensively can you fix small details and rough edges that you wouldn't have noticed otherwise. AI can build the feature, but it takes a human actually living in the software to feel what's missing.
"The best tool is the one you'll actually use. And you're far more likely to use something you built yourself, because you built it for exactly how your brain works."
We're in a strange moment. The barrier to building software is the lowest it's ever been, but most people still think of it as something Other People do. Maybe this piece convinces one person to spin up a VPS and give Claude root access. Maybe that's a terrible idea.
But probably it'll be fine.