AI agents in finance: autonomy with controls outside the model
In brief
Speakers from OpenAI, Anthropic, Two Sigma and Bloomberg at AI Engineer New York discussed letting financial AI agents act without losing control. According to OpenAI's Joseph Kim, models will soon be reliable enough for analysts to delegate more, and by 2027 agents could generate investment ideas and trade, with less Excel and PowerPoint. Others described identity, human approval, logs and stop switches around the model.
What it changes for an SME
Most SMEs will not change their week because of these talks: they concern financial services, and the 2027 trading scenario does not apply to a manufacturer or a services firm of 10 to 250 people. If you only use AI to draft text, nothing operational changes. If you plan agents that write into your tools, put identity, permissions, logs and a stop button around the agent, and keep a person on high-value actions rather than on every routine step. That is where time is actually saved and where data-protection risk is contained.
What you can do
- Write down which actions an internal AI may take alone and which need a named person's approval.
- Check that access rights, activity logs and a way to stop a run exist in your systems, not only in the chat instructions.
- Before connecting an assistant to mail, files or finance tools, ask the vendor how identity and permissions are enforced.
Let's talk about the time you could save.
In 20 minutes, we find where AI could save you time or money. Free, no commitment.