Here's What I'd Do If I Were You

Let me save you some time: if you're waiting for me to explain what a large language model is, you're reading the wrong paper. I don't care how it works any more than I care how my Bloomberg terminal renders a yield curve. I care about what it does for me by 9:30am.

Forty years. Buyside, sellside, vendors, venues. Commodities, equities, fixed income, derivatives — I've sat in every chair in the capital markets orchestra pit except maybe the French horn, and frankly I'm not sure that seat exists. I run TS Imagine. We built TradeSmart and RiskSmart X. We process trillions a week in trading volume.

I use Claude — Anthropic's AI — about ten hours a day. Not because I'm an early adopter. Because I'm lazy in the way that good traders are lazy: I want the answer faster, with less effort, so I can spend my energy on decisions that actually matter.

What follows is not a vendor pitch disguised as thought leadership. It's not a McKinsey slide deck with arrows pointing to "transformation." It's what I'd tell you at a bar after a conference day — completely unfiltered — if you asked: "What are you actually doing with this thing?"

Including the parts where I faceplanted. Those are the good parts.

Here's the dirty secret of every "AI transformation" story you've read: the hard part was never the AI. It was the data.

Most firms that try to adopt AI spend six months on what I generously call a "data engineering project" and what everyone else calls "crying into a spreadsheet at midnight trying to reconcile three different position systems that each think Tuesday was a different day."

Your market data is here. Your positions are there. Your transactions are in a third place that nobody remembers the password to. Before you can ask your first intelligent question, you've burned half a year and most of your will to live.

Unless — and this is the punchline — your data is already consolidated, current, and queryable.

The practical implication is almost comically simple: the gap between "I want to use AI" and "I'm using AI" collapses from months to minutes. No data project. No ETL pipeline meetings. No consultants with Gantt charts. You just... ask a question.

You know the ritual. Up before 6. Phone in one hand, coffee in the other, scrolling through overnight recaps, broker notes, and macro digests like a trader-shaped slot machine hoping the right information falls out.

The problem isn't that you can't find the information. The problem is that 98% of it is irrelevant to your specific book, and you're spending 45 minutes every morning panning for gold in a river of broker opinions about whether the ECB will blink.

That ritual used to take me 45 minutes to an hour. Now it takes ten. The difference isn't speed-reading. It's that I stopped reading things that don't matter.

The vendor brochure shows the trade lifecycle as a neat circle: research → analysis → execution → settlement. Lovely. Very PowerPoint. In reality, it's more like a food fight where someone occasionally asks you to produce a TCA report.

Here's where Claude actually fits — not in the idealized workflow, but in the real one. Tap any item to expand.

Let me tell you why cross-asset visibility matters more than any individual analytics tool, using an example that still gives risk managers night sweats.

It's a Tuesday morning. Headlines on Venezuela — new sanctions, political instability, the kind of rhetoric that makes diplomats nervous and traders reach for their Bloomberg chat. You're the head of trading at a multi-strategy asset manager.

The question isn't "what's happening in Venezuela?" CNBC will handle that. The question is: "Where am I exposed — across every asset class, every desk, every counterparty — right now, this minute?"

In the traditional model, answering that question is an exercise in organizational archaeology. Your commodity desk knows their crude exposure. Your FI desk knows their PDVSA bonds. Your equity desk knows their EM index weights. Your ops team has a vague sense of counterparty exposure. But nobody sees the full picture, and assembling it requires a email chain with the half-life of a congressional inquiry.

The Exposure Map — 5 Minutes, Not 5 Hours

This is what Claude builds when your data lives in Snowflake. One conversation. Every tentacle of Venezuela exposure, from heavy crude futures to the counterparty your prime broker doesn't want to talk about.

Look at what this tells you in a single glance. Venezuela isn't just the PDVSA bonds your EM desk holds. It's the crude linked to Venezuelan heavy grades. The shipping and refining equities. The EM indices where Venezuelan proxies carry weight. The counterparties with their own Venezuelan exposure that nobody thought to ask about.

Every AI article talks about alpha. None of them talk about the 60% of a head of trading's week that has nothing to do with trading and everything to do with keeping the organism alive.

I'm talking about board decks. Compliance responses. RFP answers. Vendor evaluations. The business case for trading a new asset class that your investment committee will reject the first time regardless but you have to build it anyway because that's how the game works.

This is where Claude saves you the most time, precisely because nobody has ever bothered to optimize it.

Here's the part nobody writes because they're too busy posting LinkedIn victory laps. I made mistakes with this tool. Predictable ones, humbling ones, and one that almost made it into a client presentation. I'll let you guess which.

I've organized these into three levels. Not because I'm being cute — because they represent a genuine progression in how you'll use AI on your desk. Level 1 is where everyone starts. Level 3 is where it gets interesting enough that your competitors should worry.

Every one of these assumes your data is in Snowflake via TradeSmart. Without that, you're typing context by hand like it's 2023.

But first — let me show you what I mean. Here's the exact same question asked three ways. Watch what happens to the prompt as your data infrastructure improves:

Now — the full vault. Tap a level to explore.

Before you adopt anything, know where you're starting. This isn't a test — there's no wrong answer except lying to yourself, and I can't help with that.

Right now, Claude is a tool I use. Within two years, this class of technology will be infrastructure I depend on. The distinction matters. Tools are optional. Infrastructure isn't.