- The AI types; you decide and check. Generate freely, but read the code, or have someone read it, before it touches other people’s data.
- Small steps, saved often, each briefed like a job: the goal, the limits, and when to stop.
Idea to launch in seven steps
- Plan. Who it is for, the one job it must do, and what it touches. That last answer sets how careful to be (see the end).
- Spec. Turn the plan into a brief (box below).
- Build in small steps. One feature at a time. Save each working version in version control (GitHub is one example).
- Test. Try to break it: wrong inputs, empty fields, a second user.
- Review. Read the change itself (the diff), not the AI’s summary. Did it build the right thing? Then check details.
- Deploy. Run the security checklist first. Check the host’s free plan allows your use: Vercel’s Hobby plan, for example, is for personal use only. 1
- Monitor. Once real people use it, add an error tracker, so you hear about crashes before users leave.
What to give the AI
What to check yourself
- Every package it added is real. In one large study, at least 5.2% of packages suggested by commercial models did not exist. 2
- No secrets in the code: keys, passwords, tokens. Search the version history too.
- No endless self-fixing loops. In one test, critical vulnerabilities rose 37.6% after five rounds of AI “improvement”. 3
- Logged in is not the same as allowed. Each user sees only their own data, and inputs are checked on the server.
How careful to be
Nobody’s data (prototypes, personal tools): skipping the code reading is a fair risk. Other people’s data: read the code, or have someone who can. Money, logins, personal records: not a vibe-coding job; a person who reads code reviews it.
Let the AI do the typing. Keep the plan, the checks and the final yes for yourself.
Related on this site: Lovable vs Bolt vs v0 vs Replit
- 1Vercel, Hobby plan documentation, read at source 23 Sep 2026“the Hobby plan restricts users to non-commercial, personal use only”. ↑
- 2Spracklen et al., We Have a Package for You! A Comprehensive Analysis of Package Hallucinations by Code Generating LLMs, USENIX Security 2025, read at source 23 Sep 2026 (as cited on this site’s vibe coding page): across 576,000 code samples from 16 models, “the average percentage of hallucinated packages is at least 5.2% for commercial models and 21.7% for open-source models”.↑
- 3Shukla, Joshi & Syed, Security Degradation in Iterative AI Code Generation — A Systematic Analysis of the Paradox, 2025, read at source 23 Sep 2026 (as cited on this site’s vibe coding page): “Our findings show a 37.6% increase in critical vulnerabilities after just five iterations”. One controlled experiment, not a law of every model.↑
Spracklen et al., We Have a Package for You!, USENIX Security 2025, read at source 23 Sep 2026.
The seven steps, the brief and the three levels of care are this site’s own method, drawn from the pages linked above, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK