Independent, sourced verdicts on AI-assisted software: what you can realistically build at your skill level, how to ask for it properly, and how to verify what you're given — whether it came from a model or from someone selling you a dream.
01 — The Reality Matrix
Editorial verdicts: what our own hands-on production experience says is realistic per project type and skill level. Every verdict is dated, because model capability changes — and an undated claim about AI is already misinformation.
| Project type | No technical experience | Some basics | Developer |
|---|---|---|---|
| Landing page | REALISTICDays, not hours. Expect iteration on layout and copy. | REALISTIC | REALISTICHours. |
| E-commerce store | WITH LIMITSThe storefront, yes. Payments, tax and order flows: not without review. | WITH LIMITS | REALISTIC |
| SaaS with auth + payments | UNREALISTIC ALONEAnything touching money and identity needs a security review you cannot prompt your way into.1 | WITH LIMITS | REALISTICAI accelerates; review remains yours. |
| Mobile app (published) | WITH LIMITSBuilding it: possible. Store review, signing, updates: the part tutorials skip. | WITH LIMITS | REALISTIC |
| App handling medical / financial data | UNREALISTIC ALONERegulated data. Hire a human. Non-negotiable. | UNREALISTIC ALONE | WITH LIMITSWith audit and compliance review. |
Full matrix: 9 project types × 3 levels, each cell with time estimates and the risks tutorials don't mention.
02 — Verified this week
We take the loudest claims circulating about AI coding and check them against published evidence. Three verdicts, every week. Sample below — all three are real, sourced cases.
"AI-generated code ships more defects than human code."
A December 2025 analysis of 470 open-source pull requests found AI co-authored code contained roughly 1.7× more major issues, 75% more misconfigurations, and 2.74× more security vulnerabilities than human-written code.1
SOURCE: CodeRabbit, Dec 2025 · 470 open-source PRs
"Only 0.02% of vibe-coded apps succeed."
This figure circulates widely in 2026 commentary. No methodology has been published for it, which makes it impossible to verify.2 A number without a method is a slogan, not a statistic — even when it argues our side.
SOURCE: no verifiable methodology found
"This free tutorial is neutral advice."
Multiple 2026 "beginner vibe coding" videos carry affiliate tracking links (pxf.io / Impact network) and "$499 masterclass — free" funnels in their descriptions.3 The tool recommendation is paid placement. The description exposes it; the video never says it.
SOURCE: video descriptions, on record, Feb 2026
03 — The 30-second check
Five checks, thirty seconds, no technical knowledge required. This is the shortest useful version — the full verification protocol covers functional testing, security signals, and when you are obliged to pay a human.
04 — Who writes this, and why you can check
I'm Adrian Filip. I build software with AI every working day, on real client projects with real deadlines — not demo apps for a camera. Every verdict on this site comes from that practice or from a published source you can open yourself. When I'm wrong, the correction gets published with a date on it.
Keep it independent.
This site takes no sponsors and no affiliate money — by definition, or it becomes what it documents. If it saved you time or money, you can support it directly:
What this site will never have
Sources — this page