Most AI sold into markets is vanity. As Senior Trading and Operations Associate on an OTC digital-asset desk, with 10+ years in crypto and digital asset markets and $30B+ personally traded across institutional digital-asset markets, my filter is blunt: if a product wants the button, it fails. Near the desk I trust analysis only. Timing stays human.
The first filter — does it execute, or is it vanity?
On an OTC digital-asset desk, I don’t start with model architecture, demos, or pitch decks. I start with one question: does this thing execute, or is it for vanity?
Vanity is easy to spot. It is the slide that says “AI-powered” without saying what decision it owns. It is the product that wants credit for looking smart in a meeting and never has to live with a fill, a slip, or a counterparty who does not care about your narrative.
Execute means something else. Execute means the system is asking for authority over money, inventory, or risk. That is a different grade of claim. If the answer is yes — it executes, or it wants to — I treat it as a desk risk question, not a product question. Most of what I see fails that first filter before I ever get to features.
What I’ll trust near the desk (analysis only)
Near the desk, the only AI I will trust is data analysis. Briefing. Pattern spotting. Summarizing a stack of inputs so a human can decide faster.
Nothing executable.
That is not a soft preference. It is the line. Analysis can be wrong and still leave the operator in control. Execution that is wrong leaves you with a trade you did not mean to own. I will use a model to help me see. I will not hand it the button.
As Senior Trading and Operations Associate at Stillman Digital — and first hire on that trading desk — I live in the gap between a clean chart and a live counterparty. Analysis belongs in that gap. Autopilot does not. More on that desk record is on the dossier at https://ccworldwide.ai/ and in the essays at https://ccworldwide.ai/blog/.
The hype problem — “AI algorithms” that don’t survive the operator test
The industry loves the phrase “AI algorithms.” I hear it constantly. In practice, none of them work — not when you put them next to real desk conditions.
That is my operator test, not a lab test. A lab can show a backtest. A desk has inventory, credit, timing, and a human who has to answer for the result. When the claim is that an algorithm will trade for you — or that it already knows better than the person sitting the book — it does not survive contact with how OTC digital-asset markets actually clear.
I am not arguing against tools. I am arguing against costumes. If the product cannot say, in plain language, what it analyzes and what it never touches, it is vanity with a better logo.
Human-only — when to open and when to close
When to open a trade and when to close it stays human. Full stop.
That is the last mile. Models can brief. They can surface risk. They cannot own the decision that turns analysis into exposure. I will not let a model open or close a trade for me, and I will not pretend that “AI-assisted execution” is a safe middle when the product is really asking for the button.
If builders want desks to take them seriously, they should build for the filter I use: analysis yes, execution no, open and close human-only. Sell me a better brief. Do not sell me a substitute operator.
Key takeaways
- Grade market AI with one filter first: does it execute, or is it vanity?
- Near the desk, trust analysis only — nothing executable.
- “AI algorithms” pitched as traders fail the operator test; backtests are not desk conditions.
- When to open and when to close stays human.
- Builders selling into desks should own the brief, not the button.