Why are you debugging at 11 p.m.?

Imagine opening your laptop after dinner to look at an options position that is losing money. You want to know whether the reason you entered still holds. You ask Claude to help.

An hour later, you are pasting an error message into the chat.

The script worked last week. Tonight it cannot find a column. Claude fixes that, but the chart looks strange. You ask it to check the dates. Another run, another explanation. The position is still open, but at least you have learned quite a lot about the file format.

You had intended to look at two new trades as well, and at the moment they are still on hold because... well, you are still trying to fix that script.

A losing position does not prove your original decision was bad, by the way. That is partly why you wanted to review it. But now you also need to know whether you should have trusted the analysis in the first place. And judging by the way it is handling the column in that file... you can't really shut down the little voice in your head whispering that maybe trading with Claude was a bad idea.

And you would do well to listen to that voice.

Claude can help an engineer build a good research system. It can write code, help develop models and work through tests. If you want to own that system, having an assistant beside you is useful. But you cannot simply hand Claude responsibility for being the architect, the builder, the maintainer, the auditor, the trader and the risk manager.

And no, let's be straight: opening multiple instances of Claude does not settle that either.

Maintaining the data and validating the models can be a full-time job. That is why you have so many people in a trading firm. If all you needed were a few PhDs, what would all those other people be doing?

The work is divided between traders, quants and other specialists, though the roles can overlap. And even within the quant roles, you will find a wide variety of responsibilities, with each person carefully working on a part of the chain that helps produce good decisions.

Yes, literally: imagine a factory, Ford style, where a quant developer is in charge of maintaining APIs, a data engineering team is in charge of maintaining specific datasets, and risk managers and traders are constantly in touch to analyse the feedback they get from the market.

And now there is you. And Claude.

Feels a little lonely all of a sudden, doesn't it?

It shouldn't. And truth be told, while these companies set a high standard, not everyone is in a position to compete with them. It doesn't mean you can't take inspiration from them.

So what can you do, or even better, what should you do to get there?

Start by investing in a data intelligence layer. Stop trying to build it and maintain it with Claude. We get it, vibe coding has made a lot of things possible, but would you really try to send a rocket into space... from your office?

Okay, investing your money isn't rocket science... but it still demands rigour and patience. To put it simply: it is a real job. You can try to do it yourself, or you can trust people who have been doing it for a long time.

What does that look like? Your assistant asks our API for volatility history, put-versus-call skew, or volatility risk premium and gets the results. We maintain the calculations behind those metrics. You aren't back fixing the scripts that produce them. One of those terms wasn't on your radar? Ask Claude what skew says about puts versus calls, or what volatility risk premium says about the gap between implied and realised volatility, and whether either one matters for your position.

Then there are forecasts. Descriptive analytics tell you it is sunny today. Useful, but what about tomorrow? Especially in a country where the wind turns quickly. By the way, isn't that ... exactly what the markets do all the time? Many people are still stuck looking at a metric because it worked historically. But that gives you little visibility into whether it is still relevant to the setup sitting in front of you. Our forecasts give your assistant an estimate to examine before you commit money to a trade just because a perfectly optimized backtest said so.

And now, all of a sudden, your assistant, whether it's Claude, Kimi or Astra, has more to work with when you ask it to help you make a decision. Is implied volatility in the ticker you're eyeing likely to exceed the subsequent realised volatility over the same period? If yes, that could be the basis for investigating a trade. If not, your assistant has evidence to help you consider staying away or looking for a better setup, instead of agreeing with you by hallucinating a plausible-sounding thesis with no real grounding.

And ultimately, you and your assistant are better equipped to make good decisions.

See! Look at you getting closer to how the big boys work, one trade at a time!

Life is a collection of decisions, and trading is not any different. If you want to succeed, the quality of those decisions matters, including what you stand to win or lose. The first one that could change your trading experience is connecting your agent to our API and getting access to our descriptive and, most importantly, predictive analytics. You'll need an API key and a connection set up for your agent; our MCP server is coming soon.

That is our job. We do it well, so that you can focus on the trading thesis.