Goldman just pulled the plug on trader coding.
Back in July, their CIO was hyping "vibe coding" — traders spinning up their own tools with Claude and GitHub Copilot. Flow monitors, position models, quick text and graphics work. Engineers would clean it up later.
Then October happened. After a trial weighing productivity against risk, Goldman restricted those tools to engineers only. Strats — the quant-trader-engineer hybrids — got a pass. Regular traders did not. Some are upset.
Goldman's line: "We're keeping the tools with employees best suited for longer use."
Translation: traders can prototype, but they can't own the build.
The real question isn't about access. It's about momentum. How much of that trader-built workflow actually survives once only engineers can touch it? How many of those scrappy internal tools just... stop getting updated?
This is the tension every firm will face. AI lowers the barrier to building. But production code still needs guardrails, version control, and someone who knows what they're doing when it breaks.
Goldman chose control over speed. That's probably the right call for a bank. But it's also a bet that the engineer queue won't become the bottleneck.
Back in July, their CIO was hyping "vibe coding" — traders spinning up their own tools with Claude and GitHub Copilot. Flow monitors, position models, quick text and graphics work. Engineers would clean it up later.
Then October happened. After a trial weighing productivity against risk, Goldman restricted those tools to engineers only. Strats — the quant-trader-engineer hybrids — got a pass. Regular traders did not. Some are upset.
Goldman's line: "We're keeping the tools with employees best suited for longer use."
Translation: traders can prototype, but they can't own the build.
The real question isn't about access. It's about momentum. How much of that trader-built workflow actually survives once only engineers can touch it? How many of those scrappy internal tools just... stop getting updated?
This is the tension every firm will face. AI lowers the barrier to building. But production code still needs guardrails, version control, and someone who knows what they're doing when it breaks.
Goldman chose control over speed. That's probably the right call for a bank. But it's also a bet that the engineer queue won't become the bottleneck.