David Tacer (@TacerDavid) header
David Tacer (@TacerDavid) profile picture
Somewhere around the blocks
Joined April 2013
Photo by David Tacer (@TacerDavid)Photo by David Tacer (@TacerDavid)Photo by David Tacer (@TacerDavid)Photo by David Tacer (@TacerDavid)Photo by David Tacer (@TacerDavid)Photo by David Tacer (@TacerDavid)Photo by David Tacer (@TacerDavid)Photo by David Tacer (@TacerDavid)
Rule I follow with AI agents: After 2–3 failed attempts on the same bug, stop. Stash the changes. New session. Better evidence. Persistence helps humans. It confuses agents.
4
Write every piece of software as if you can’t patch it later. Smart contracts taught me that. It’s a good habit everywhere.
4
Parents: what's the #1 thing that makes you trust (or delete) an app your kid uses?
3
Tests aren't for engineers only. A test is just: "when I do X, Y should happen", written down so the computer checks it every time. Ask your agent to write one for every bug it fixes. The bug never comes back quietly.
7
Your deploy failed. Don't scroll to the bottom of the log. Scroll to the FIRST error. Run the build locally. Give that exact error to your agent. Ask for the cause before the fix.
12
Pushed your API key to GitHub? Deleting the commit does NOT un-leak it. Rotate the key in the provider's dashboard. Now. Then move it to an environment variable.
22
You don't need to learn to code to build with AI. You need to learn to: – define what done means – read a plan – test like a user – check what changed – ship safely That's the job now.
7
The #1 security hole I see in AI-built apps 🧵
2
25
5/6 Fix: turn on RLS for every table and write policies like "users can only read rows where user_id = their id". Then test with a second account.
1
6
6/6 Ask your agent: "List every table and its RLS policies. Which ones allow access to other users' data?"
3
Before you start a feature, write one sentence: "The user can ___ so that ___." If you can't write it, the AI can't build it.
11
Kids promise to walk the dog. Parents end up doing it. I'm building PetPrep: a 12-week AI pet simulation with real steps, real feeding windows and a live dashboard for parents. Objective proof before a real pet comes home. petprep.si
1
28
The AI said "Done! Everything works." It didn't test it. "Done" means you clicked through it yourself. On your phone. As a new user. With bad input. Trust, but test.
1
16
When your app breaks, don't tell the AI "it doesn't work". Tell it: – what you did – what you expected – what you saw (exact error) – where (page, screen size) Then: "Find the root cause. Don't change anything yet." Evidence first. Fix last.
9
If you build with AI and don't use git, you're working without an undo button. 3 commands are enough to start: git status git diff git commit Know what changed. Save checkpoints. Go back when it breaks.
7