Research
Bloomberg's head of market structure research says trading desks are closer to real agentic AI than at any point in four decades of false starts, but the honest read is that the first wave is workflow augmentation, not autonomous execution
Larry Tabb told Traders Magazine that despite repeated attempts at desk automation since the late 1980s, current agentic AI is the closest the industry has come, while cautioning that "the promise of Agentic is large, but the reality is uncertain" and that infrastructure has to be built before automation can expand.
Tabb's framing is useful precisely because it resists the two easy narratives that dominate agentic AI coverage: that autonomous trading agents are already here, or that agentic AI on the desk is hype that will fade like the automation attempts before it. His actual position, that this is genuinely the closest the industry has gotten in roughly four decades of trying, while still describing the near-term reality as uncertain, is a harder claim to make than either extreme and a more useful one for anyone trying to plan around where this technology actually is.
The specific target for the first wave, repetitive and information-heavy work that consumes much of a trader's day, is a deliberately narrow and defensible starting point. That is the category of task where an agent's failure mode is annoying rather than dangerous: a summarisation error or a missed data point in pre-trade research costs time to catch, not capital. Starting there rather than at execution is the same caution implicit in Tradeweb's public concern this week about rogue AI agents on live venues, just arrived at from the buy-side research angle instead of the market infrastructure angle.
Why infrastructure has to come first is the part easy to skip past but worth sitting with. An agent that can summarise research or flag anomalies is only as useful as the data pipes, permissioning and audit trail underneath it, and building that layer properly is unglamorous work that doesn't generate its own headlines the way a flashy autonomous-trading demo does. Firms that skip straight to agent deployment without that foundation are the ones most likely to produce an agent that looks capable in a demo and behaves unpredictably in production, which is exactly the gap Tabb's caution is pointing at.
Read alongside this week's other coverage, the pattern across market-structure research, venue operators and asset managers is unusually consistent for a technology this new: augment the decision-support layer first, keep humans holding execution authority, and treat full autonomy as a later-stage goal rather than a current capability. That is a notably more conservative consensus than the public conversation about AI agents in finance usually conveys, and it is coming from the people closest to actually building the systems rather than commentating on them from outside.
Read the original: Traders Magazine - Bloomberg's head of market structure research says trading desks are closer to real agentic AI than at any point in four decades of false starts, but the honest read is that the first wave is workflow augmentation, not autonomous execution. Commentary is the independent editorial view of Share Trading; the original article is credited to its publisher.