There's a big difference between someone who collects tools and someone who operates them. The collector subscribes to twenty and runs none. The operator knows theirs by name, and knows which one leaves without pain.
Inventory is one of the five pillars of the thesis I follow, and not for decoration. A stack is not a showcase of logos in a portfolio footer. It's the list of what you have, what each piece does, and what it costs when the project grows.
The five pieces that open the door
To get out of zero and have something actually live, five are enough.
GitHub is the plot of land with the deed on record. Your code lives there and every change gets logged with date and author. Broke something badly? Roll back to before the mistake. Free, even for private repos. I suggest GitHub Desktop, it makes life easier.
Vercel publishes the site with no server for you to babysit. git push and the house is live.
Convex is database and backend in one place: the data changes, the screen updates by itself, no taping three services together.
Porkbun registers your domain cheap, with no renewal trap. The address stays in your name, not in a platform's.
Claude Code builds with you. It's the shortest path between the idea in your head and the thing running. GPT's Codex is good too.
Monthly cost of the whole set: close to zero. What stalls a project at this stage is not the stack, it's the decision to start.
Helen learning the lean operator's mindset
The rest arrives with the pain
The temptation is to furnish the whole house before living in it. You subscribe to the email tool, the payments tool, the analytics tool, the scheduling tool, and spend the month configuring instead of publishing.
Charging someone? Stripe. Email landing in spam? Resend. Video weighing down the database? Bunny. Need synthetic voice? ElevenLabs. Meetings turning into hand-typed minutes? Granola. Posting the same thing on five networks, one by one? Postiz.
The five essential pieces of a digital stack
The four filters
Before any piece gets in, it answers four questions.
Generous for free. The free plan handles real small-scale production, not just a weekend test.
Disappears into the routine. Managed, with an API or MCP, so nobody wakes up in the middle of the night to restart a server.
Price behind the usage. The bill grows after the project grows, never before.
Exit door open. Easy export, open standards. A tool that takes hostages doesn't get in.
A piece that fails two of these filters can still get in, as long as you know which one it failed and what will hurt later.
Where this ruler cuts
The fourth filter is the easiest to write and the hardest to honor. Managed services quietly charge a piece of your freedom: whoever hands infrastructure to someone else to run is betting that company will still exist in three years, at the same price.
My working assumption is that the bet holds as long as the data walks out the door intact. A database with export is reversible, even if it hurts. A workflow locked in a panel that only exists inside one platform is not.
An honest hole: this is a five-piece list that works well for the web. If you're going native app, hardware or your own model, you'll swap half of it in the first month. Especially early on, don't be afraid to start over. It pays off in the long run.
Open the list and count
Worth more than any recommendation: today, list everything you pay for or use in the project, including what you forgot you subscribed to. Note what each piece does. Run them one by one through the four filters and write down where each failed.
Then pick one to cut or replace this week. Canceling is also building.
If a hole shows up in the list, the AItag catalog is where I go looking for the piece before subscribing on impulse. Every tool there went through hands-on testing, not a press release.
How many pieces do you have today that would not survive the exit-door filter?
