The word “expert” got cheap. A weekend of prompts, a viral thread, a conference badge — and suddenly the industry has a thousand AI experts. Expert and novice are not personality types. They are different piles of receipts.
I care about the difference because novices with microphones set strategy for companies that cannot afford a costume. Owners deserve a cleaner test than charisma.
What is the difference?
A novice can describe the tools. An expert can describe the failure modes, the infrastructure, the evals, and the human loop when the model is wrong. A novice ships a demo. An expert ships something that survives Tuesday — demo ≠ product.
A novice talks models. An expert talks architecture — same models, different outcomes. A novice sells certainty. An expert publishes dates — why dates matter.
When the label gets dangerous
When a novice designs your stack. When a novice trains your team into compliance instead of curiosity (Training Inversion). When a novice takes your association’s mainstage and teaches fear or fantasy — don’t take the mic if you’re not qualified.
The room is often ahead of the influencers anyway — I wrote that gap. Contractors prototyping in the shop can smell a novice in the first ten minutes. Program committees sometimes cannot.
Failure modes of fake expertise
- Logo fluency. They can name every model and cannot answer one infrastructure question — behind the curtain.
- Retainer cosplay. They sell AI as a monthly fog — the same invisible-work pattern as broken SEO retainers ( agency model is broken).
- Borrowed war stories. Every case study sounds like a keynote they attended, not a system they run.
Proof — how I want to be tested
Ask me what we run. Ask about the rebuild numbers — 320 to 38. Ask what we got wrong and when we will review the call (Prediction Ledger). Ask for the curtain: data flow, permissions, evals, who owns architecture. If I cannot answer, I am not your expert — I am a storyteller. Hold every other “expert” to the same standard.
A ninety-second interview script
"Show me something you run in production that breaks when the model is wrong — and how you catch it." "What did you get wrong last quarter, and when will you review that call in public?" "If I ask one infrastructure question, will you answer without a slide?" Novices stall. Operators light up. You do not need a procurement department to hear the difference.
Apply the same script to yourself before you teach. If you cannot pass it, you are a student with a microphone — and students should not set industry strategy from mainstages or retainer decks.
Titles will keep getting louder. "Chief AI Officer." "AI Evangelist." "Prompt Engineer of the Year." Measure the work under the title. Production systems. Dated misses. Open curtains. That is the difference — and it is available to a contractor owner as much as to an agency CEO.
When you find a real expert, you will feel the opposite of a pitch: fewer promises, more constraints, clearer ownership, and an invitation to verify. That posture is rare. Prefer it over certainty every time — especially when the certainty is for sale.
Pair this essay with Make Them Show You Behind the Curtain before every hire, and with Don't Take the Mic before every speaker booking. Expertise is a cluster of tests — not a compliment you award for confidence.
Action
Before you hire a partner or book a speaker, run two tests: one infrastructure question with a demand to see behind the curtain, and one request for a dated claim they will let you score. If you want an operator on your stage, start booking. If you want the team that builds the layer, Work with Jennifer. If you want the literacy without the hire pitch, read AI is not ChatGPT and tools are not agentic systems.