After decades of electronification and automation, the next stage in the evolution of the buy-side trading desk may be less about removing traders from execution than changing the information, tools and decisions that reach them.
That was one of the clearest themes to emerge from A-Team Group’s recent TradingTech Insight webinar, The Buy-side Trading Desk of the Future, sponsored by LSEG Data & Analytics. Bringing together Dwayne Middleton, Global Head of Fixed Income Trading at T. Rowe Price; Stuart Lawrence, Head of European Equities Trading at UBS Asset Management; and Holden Sibley, Head of Buy-Side Community at LSEG Data & Analytics, the discussion explored the convergence of trading infrastructure across asset classes and the growing role of AI within execution workflows.
Technology converges faster than markets
Thirty years of standardisation have created considerable commonality across trading technology. OMS and EMS platforms, FIX connectivity, electronic execution and analytics increasingly provide a shared functional framework across equities, FX and fixed income.
But the underlying markets remain very different.
“While the infrastructure is converging, technology is converging faster than markets currently are,” said Lawrence. “Equities are very much traded now on liquid markets, whether it be exchanges or via SIs, pools, multilateral venues, etc., while FX remains an OTC streaming RFQ approach, and then with fixed income that’s even more heterogeneous.”
Middleton pointed to a similar distinction within fixed income itself. Electronic trading, portfolio trading and auto-execution have made parts of investment-grade credit increasingly resemble equities, while less liquid instruments still depend heavily on relationships and dealer interaction for price and liquidity discovery.
That helps explain why trading architectures remain only partially integrated. An audience poll found that 54% of respondents operated some shared infrastructure while retaining significant asset-class-specific systems. Just 15% described their architecture as fully integrated.
Connecting rather than replacing
The implication is that the future multi-asset desk doesn’t necessarily require every execution workflow to be consolidated onto a single platform. A common data foundation combined with greater interoperability may be more important.
Middleton described the objective as creating a “single pane of glass” for the trader, where the applications operating behind the interface become less relevant and information relating to an order is pushed to the trader when required.
Sibley sees APIs becoming increasingly important in creating that environment.
“The actual execution layer caters to really different market structure characteristics and nuances across asset classes, so it wouldn’t make sense to try to unify that execution layer,” he said. “Having a consolidated book of records beneath that is a critical piece, but over top of those asset-class execution layers is where the opportunity lies – analytics, collaboration, UI.”
That architecture also provides a foundation for AI to operate across previously separate workflows. Rich APIs can connect specialised systems while allowing an AI layer to draw information from across the trading environment without requiring firms to undertake wholesale platform replacement.
AI moves closer to execution
AI has already made significant inroads into research and idea generation. Execution presents a harder problem because decisions are time-sensitive, information is incomplete and mistakes can have immediate financial and regulatory consequences.
The near-term opportunity therefore lies largely in augmenting decisions rather than making them autonomously.
Middleton outlined how an AI agent could bring together liquidity, pricing, previous trading history, dealer interactions, research and news around an individual order, allowing the trader to concentrate on the decision rather than gathering the information required to make it.
In fixed income, that could begin even before an order exists. AI could help determine whether an investment idea is realistically executable by bringing liquidity into the conversation between analysts, portfolio managers and traders earlier. It could also streamline data gathering and validation around new bond issues.
Lawrence envisaged how a similar model could work in equities: “If I get a tricky order, I click on the order and AI brings up my entire trading history with that, including brokers previously used, broker performance, who mentioned it recently on Bloomberg chat or email – giving me information that usually would take 30 or 40 clicks literally by putting my cursor over the order.”
An augmented trader
That distinction between supporting a decision and taking responsibility for it is likely to shape AI adoption on trading desks.
“What people mistake is the concept that it either has to be full autonomy or nothing,” said Lawrence. “The key is transformation: the future trader is AI-augmented, not replaced.”
He expects AI to take on increasing amounts of information gathering, liquidity analysis and risk profiling. But unusual or stressed markets expose the limitations of models trained largely on existing patterns.
“On a normal day in calm seas, AI can probably do 20 to 30% of my job in the next two to three years in terms of trading,” he said. “When we have situations with no precedent or stress situations in the market, AI is going to struggle, and that’s when human triage needs to kick in very quickly.”
Accountability presents another boundary. Middleton noted that clients appoint asset managers because they are entrusting human investment professionals with their money. Explaining that an AI agent was responsible when something went wrong would be a difficult conversation.
Building the foundation
The changing division of labour will also affect the skills required on trading desks. As natural-language tools make it easier to create applications and interrogate data, coding ability may become less important than critical thinking, cross-asset knowledge and the ability to understand risk.
Middleton also highlighted a potential challenge for the traditional apprenticeship model if junior traders no longer learn by performing some of the lower-touch tasks increasingly handled through automation.
For firms preparing for that transition, however, the panel returned repeatedly to something more fundamental than AI models themselves: data.
Asked what a buy-side COO should prioritise over the next 12 months, all three panellists put strong data foundations at or near the top of the list. Lawrence stressed the importance of cleansing and preparing data before deploying agentic tools, while Middleton advocated starting with simple, practical applications and retaining enough flexibility to adapt as models evolve.
For the buy-side trading desk of the future, the immediate task is therefore less about achieving autonomy than creating an environment in which AI can be trusted to take on more of the work surrounding the execution decision. As that environment develops, the trader remains at its centre – but with considerably more technology working on their behalf.
A recording of this webinar is available at https://a-teaminsight.com/webinars/the-buy-side-trading-desk-of-the-future/?brand=tti
Subscribe to our newsletter
