Goldman gates its coding agents; a veteran ML engineer quits LLMs outright

thestateofai.com · yesterday

Goldman cuts its traders off from coding agents

Takeaway: roughly 12,000 of its 47,000 staff

A quiet Sunday, and one story is worth the coffee. Per Bloomberg News, reported November 8: Goldman Sachs decided in October that only its engineers — roughly 12,000 of its 47,000 staff, by its CIO's count — may keep using AI coding agents such as Anthropic's Claude Code. Bankers and traders, who had been building tools with them to track client activity and model positions, lost access after a review covering both risk and productivity.

This is a first big bank saying openly that code written by untrained hands with an agent is a regulated-business liability, not a productivity win — bad for agent tools sold on everyone-gets-a-seat, good for gated, audited enterprise deployments. The irony is sharp: Goldman remains a lead underwriter on Anthropic's IPO, per the S-1 filed October 26, and its own bankers will market a product they can no longer run. Ask every founder you back: who owns a script a marketer or analyst builds with an agent before it touches customer data?

Read it at thestateofai.com

polso.info · yesterday

Kristian Polso

On November 8, Finnish developer Kristian Polso — about ten years building production ML systems for clients — wrote that he will not use LLMs or coding agents in his professional work at all. His reasons run from the mental fog heavy users report, to agent-written code that looks polished but is costly to maintain, to plain love of the craft, and, as an EU citizen, to US providers sitting under the CLOUD Act, which lets US federal law enforcement reach data stored anywhere in the world.

The received wisdom says engineer adoption of agents is universal and inevitable; here is a named, credible practitioner saying no, echoing a quieter current among senior developers. For the companies you back, that cuts two ways: some of the strongest hires will refuse the toolchain, and European buyers may ask harder questions about where their code and data live. Ask your founders: who reviews what the agents write, and would the team even notice if a senior engineer quietly stopped using them?

Read it at polso.info

“"AI is here to stay." OK. So is heroin. And I wouldn't recommend that to anyone either.”