Traders Building Their Own Tools Is the New Excel. Don't Ban It, Give It a Path to Production.

Goldman Sachs has limited Claude Code and GitHub's AI coding tools to its engineers, Bloomberg reported on 2 November 2026. For months, traders and bankers had been using them too: to monitor client flows, model market moves and positions, and put together written and visual material. Now access sits with the firm's roughly 12,000 engineers, and reportedly the strats who work across trading, quant research and engineering.

Our first reaction was surprise. We completely understand the reasoning, but we don't think a full block is the answer.

Traders Have Always Built Their Own Tools

First Excel, then Jupyter notebooks for the more technical ones. On every trading screen we've built, someone asked for "export to Excel". The moment traders have the data, they want to work with it their own way.

They do it to unblock themselves. Getting an engineering team to build something means process: requests, prioritisation conversations, waiting your turn. And often they don't know exactly what they need until they've tried something. Building it themselves is how they find out.

What's changed is that these experiments are now usable enough to be products. With an AI coding tool, a trader can build in an afternoon what used to be a rough sketch handed to engineers. That's powerful, and it's also where the trouble starts.

The Risks Are Real

Distraction and context switching. This is the big one. People have defined roles for a reason. Anyone can learn and train to get good at other roles, and at a firm like Goldman many people do. But if everyone goes in every direction, doing everything halfway, nobody goes in the right direction full on. Every switch between watching the market and fixing a tool breaks focus.

Drifting away from engineers. When building alone feels faster, the slower shared route stops feeling worth it. That means fewer conversations, less shared understanding, and more tools only one person knows how to run.

Chaos. Many people build many tools, often the same ones. The result is duplicated tooling, rising token bills and nobody with the full picture.

Data leak. We suspect big firms have the approved, integrated AI tools well covered. But there are countless other AI tools out there, and nothing really stops someone from signing up to one because it may turn their data into a cool presentation.

And two costs nobody planned for: maintenance and bugs. A tool is never "done". Someone has to maintain it, improve it, deploy it and connect it to more things. That's why engineering teams have ever-growing backlogs. And code that hasn't been reviewed or tested can be buggy in ways nobody notices, until a wrong number on a screen leads to a wrong trade.

The Upside Is Just as Real

Distraction has a cost, but so does not experimenting.

Speed. Traders need to move fast, and markets don't wait for a sprint. With these tools a trader can test an idea the same day, instead of writing a request, having meetings and prioritisation discussions, and waiting weeks to find out it wasn't quite right.

Unblocking themselves. They solve their own small problems without queuing for engineering time. That frees engineers for the work only they can do.

Ideas. More people experimenting means more ideas, and some of the best ones come from the people closest to the market. Goldman's own CIO praised exactly this in July: non-technical staff building with AI, then handing it to engineers to refine.

Building the right thing. This is the one we like most. Often traders don't know exactly what they need until they've tried something. After making a rough version themselves, they know far better what they want. The conversation with engineers starts from "this, but solid" instead of a vague request, so less gets built wrong. A trader-built prototype is the best spec an engineer can get.

Signals, not just noise. Even lots of people building similar tools tells you something: if five people on a desk built the same thing, that's a real need. With some visibility, that's a ready-made roadmap.

Better partners. Traders who have built something understand the trade-offs better. They ask sharper questions, and working with engineers gets easier, not harder.

What We'd Do Instead: Treat It Like the New Excel

It's Excel all over again, only far more powerful. Excel didn't replace engineering teams; it made traders better at their jobs and showed engineers what to build next. This can do the same.

Traders build to experiment, inside the firm's approved tools, with no sensitive data leaving them.

Anything that becomes useful goes to engineers to productionise and support.

Managers and MDs know what their people are building, through a shared list or a regular demo.

They collect the common needs across the desk and ask engineers to build one proper version to replace the many.

Traders keep their speed, engineers get clear direction, and the firm gets tools it can rely on.

We hope and expect that Goldman finds a way to give traders access back with the right guardrails. The answer isn't fewer people building. It's communication, visibility, and empowering both sides.