I heard about this concept first on a podcast episode of the Rare Mind podcast by strategy consultant Alex H Smith, interviewing fellow consultant Katelyn Bourgoin. The relevant bit comes up in the first ten seconds of the episode: You can talk to Claude for hours (or months) about a tricky topic and not get anywhere, because you don't know if you can trust it. Thus, you make much faster headway by working with a human expert.
I experienced this firsthand with a non-technical founder who contacted me for a mini version of our Blueprint service. The founder, non-technical but with an ambitious and worthy idea, had spent quite some time vibe-coding a particular app. Knowing these vibe-coded apps often have security holes, he took the app (generated by OpenAI's Codex) and asked Claude to give a second opinion. Sure enough, Claude found some issues that he then had Codex fix. Still not satisfied, he passed the whole thing to Grok to find even more issues and have those fixed too. Alas, he still wasn't satisfied, and that's why he reached out to me: To have a human expert give a technical review, identify gaps, and come up with a plan for what can stand as is, what needs to be rebuilt, etc.
Nothing's been signed yet, so I can't say for sure whether the founder recognizes the value in an expert review. What's true, though, is that he's hesitating and doesn't feel confident about releasing the app, approaching investors, or approaching potential business partners for a pilot.
It's interesting: Claude and co are great and diligently combing through a codebase. If we get retained to do the blueprint project, I'll definitely rely on Claude to make sense of the codebase. Why then pay me to do it? Simple:
I know what questions to ask, in a more pointed way than "make me an app and make it secure"
AI agents are over-eager and try so hard to be helpful that they'll most certainly raise security issues that are purely theoretical, or don't apply during the intended use (AI coding agents are prone to over-engineer and be overly defensive)
My judgment doesn't randomly flip from one thing to another just because of subtle differences in how you asked me to do the thing ;)
My judgment also depends on a lot of "context" and experience that a non-expert can't hope to correctly provide in their prompt
No amount of Codex, Claude, and Grok pinky-promise swearing that the app is secure will give you the confidence you need in the way a thorough human-led (if AI-empowered) review does.
PS: If you know someone - friends, colleagues, family members - who have great ideas about what they'd want to build, but lack the confidence to jump in, or are overwhelmed with the options, or have concerns about security, send them my way. I'd love to learn more about the specific challenges.
