Vibe coding makes AI experts hard to trust: how to hire anyway

Many of my clients hire AI experts or developers. The candidates show impressive portfolios, but many of them crumble when the client asks them to debug a complex scenario. Here is what 15 years of building software taught me about this: the finished product no longer proves skill.

AI tools make polished work cheap, so buyers can no longer tell a real AI expert from a vibe coder by looking at their portfolio. That turns AI expertise into a lemon market. To hire well, stop judging the finished product and test the process: ask candidates to explain how they solved a real problem and why.

Vibe coding: expert-looking work without structure

Vibe coding is when people use AI to generate code, content or strategy that feels right and looks professional but lacks structural integrity. The output passes a quick look. It fails when someone asks the author to debug a complex scenario or explain the architecture.

A case from December 2025 shows how far this goes. The Guardian reported that a single author, Kevin Zhu, claimed 113 AI research papers in one year, many co-written with the high school students his company mentors. Berkeley professor Hany Farid called the papers a "disaster" and said he was "fairly convinced that the whole thing, top to bottom, is just vibe coding." That is vibe coding applied to science, and it shows how easy it is to fake expertise.

Why AI expertise is becoming a lemon market

In economics, a lemon market happens when the buyer cannot tell a high-quality product from a low-quality one (a "lemon"). George Akerlof described this problem in 1970, and the work later earned him a share of the Nobel Prize. When buyers cannot tell the difference, trust evaporates and the market collapses.

The market for AI expertise is moving in that direction. Credentials and portfolios are becoming unreliable signals, because AI can produce both at volume. A good-looking portfolio costs a vibe coder almost nothing. A real builder has to earn the same result. From the outside, the two portfolios look the same.

How vibe coders fail in real projects

I see the same pattern with clients who hire AI experts or developers. The candidates present impressive portfolios, mostly generated with ChatGPT. Then the client asks them to debug a complex scenario or to explain the architecture, and they crumble.

These candidates are vibe coders. They know how to prompt, not how to build. If you want to see how to design a system before anyone builds it, the Automation Architecture Guide shows the patterns a real builder should be able to name.

How to hire AI experts you can trust

The finished product is now cheap, so stop looking at it first. Look at the process. Use these steps in your next interview:

1. Ignore the portfolio volume. A long list of polished projects proves nothing if AI made them. Treat the portfolio as a list of topics to ask about, not as proof.
2. Ask how, not what. Pick one project. Ask the candidate to walk you through how they solved the problem, step by step.
3. Test the architecture. Ask why they chose this structure and what breaks if one part fails. A builder can answer. A vibe coder repeats what the tool produced.
4. Give them a complex scenario to debug. Use a real problem from your business. Watch how they find the cause, not only whether they fix it.
5. Ask for proof of their process. Ask for notes, diagrams or documentation from a past project. These show whether they understand what they build.

If a candidate cannot explain the logic, they are renting intelligence from the AI, not owning it. For a wider view on what to test when you hire in the AI era, read why you should hire for judgment, not AI skills.

The bottom line

AI made expert-looking work cheap, and that broke the old way to judge expertise. Do not hire for the output. Hire for the person who can explain, defend and fix the output.

Build a calmer business,Yuri

Frequently asked questions

What is vibe coding?

Vibe coding is when people use AI to generate code, content or strategy that feels right and looks professional but lacks structural integrity. The output passes a quick look. It fails when someone asks the author to debug a complex scenario or explain the architecture, because the author knows how to prompt but not how to build.

What is a lemon market for AI expertise?

A lemon market happens when buyers cannot tell a high-quality product from a low-quality one. Trust then evaporates and the market collapses. AI expertise is moving in this direction: AI tools make polished portfolios and papers cheap, so a business that hires AI consultants or developers can no longer judge skill from the finished work alone.

How do I check if an AI consultant or developer really knows what they are doing?

Ignore the volume of the portfolio and test the process. Ask the candidate to walk you through how they solved one real problem, not what they built. Then give them a complex scenario to debug or ask them to explain the architecture. A candidate who cannot explain the logic is renting intelligence from the AI, not owning it.

Why are portfolios and credentials less reliable for hiring AI experts?

AI tools make finished work cheap to produce. In December 2025, The Guardian reported that one author, Kevin Zhu, claimed 113 AI research papers in one year, many co-written with high school students. Berkeley professor Hany Farid called the papers a disaster and said he suspected the whole thing was vibe coding. If research papers can be produced at that volume, a polished AI portfolio proves little about real skill.