Ask most trading desks what slows them down and they will point to the months between deciding to trade a new asset class and being allowed to, the client channel they cannot use because nobody can supervise it, or the market they entered late because the controls were not ready. That delay is not simply a compliance problem. It is a commercial one, and it is measurable.

This session will explore how controls can become the fastest route into a new market rather than the obstacle in front of it. Discover how defining risk once and applying it consistently across channels, products and jurisdictions can turn expansion into a configuration exercise rather than a lengthy approval process, giving firms earlier market entry, more channels, wider coverage and less time spent asking permission to move.

AI has stopped recommending and started acting. Agents are moving into investment and trading workflows faster than most firms have built the means to govern them, and the instinct is to slow adoption until the governance catches up. That instinct is expensive and usually wrong.

The firms making real progress have inverted the question. Instead of asking how much governance they can afford before it costs them speed, they are asking what governance buys them: the ability to put an agent into production and defend it afterwards, to give the desk a new capability without a six-month review, and to answer what happened and why without assembling a project to find out.

This working session looks at how to build that in practice. Where the line sits between an agent that recommends and an agent that acts, and what has to be true before you move it. Where oversight genuinely belongs in the workflow and where it only appears to. How to raise risk awareness across a desk without adding a click to every trade. And what a defensible record of an autonomous decision actually contains, when the regulator asks six months later.

We will use a live example: AI agents already running first-line review at institutional scale, where three independent models must agree before anything closes autonomously and any single dissent escalates to a human. Not as a product demonstration, but as a worked case of what it takes to let something act on your behalf and still be able to prove the control worked.

The commercial argument runs underneath all of it. Every one of these controls looks like a constraint and behaves like an accelerator. A firm that can evidence what its agents did is a firm that can deploy the next one without a six-month review, put the desk on a new channel without banning it first, and enter a new market on schedule rather than on appeal. Governance is not the tax you pay for AI on the trading desk. It is the reason you get to use it at all.

Participants should come with the decision they are currently unwilling to automate. We will work through what would have to be true to change that.