The trading industry has spent years refining execution algorithms, automating workflows and feeding ever-larger volumes of market data into increasingly sophisticated systems. Now attention is shifting to a new generation of technology: agentic AI. Unlike traditional algorithms that follow predefined rules, these systems promise to interpret context, adapt to changing market conditions and take action toward specific objectives.
But despite the excitement, industry experts say the technology is still finding its place.
“The distinction comes down to whether the system is simply executing a better set of instructions, or whether it can operate across a workflow with context, adaptation and governed autonomy,” Mehmet Kinak, Global Head of Equity Trading at T. Rowe Price, told Traders Magazine.
Traditional execution algorithms remain highly effective, Kinak said, but they generally operate within fixed parameters. An agentic system, by contrast, would need to “observe changing market conditions, reason across multiple sources of information, decide when to act or not act, and document why it made a recommendation.”
That does not require fully autonomous trading, he added. Instead, the technology should help traders synthesize market color, liquidity signals, order context and execution analytics while keeping humans accountable for the final decision.
Audrey Costabile, Senior Analyst, Market Structure & Technology at Crisil Coalition Greenwich, also sees adaptability as the defining characteristic separating agentic AI from conventional algorithms. “Functionality which is more adaptive—for instance, an algo that can react and reconfigure based on market movements and other information… are beginning to emerge,” she said. More advanced ETF algorithms that incorporate information about both the fund and its underlying securities are already demonstrating how execution technology is becoming increasingly dynamic, she said.
Data Comes Before Intelligence
For fixed income trading, Dwayne Middleton, Global Head of Fixed Income Trading at T. Rowe Price, argues that the industry’s biggest challenge is not intelligence but infrastructure.
“A more advanced algorithm still executes rules,” Middleton said. “A genuine evolution pursues an objective instead: it takes a goal inside defined limits, works through the steps, and hands control back when conditions move beyond its mandate.”
That distinction has become increasingly relevant as fixed income trading shifts from identifying individual securities toward constructing desired risk exposures, he said.
“The desk now defines a target risk profile across duration, spread, sector, rating, seniority, and liquidity, then sources the liquidity that fits it,” he said.
Making that possible depends on connecting fragmented systems, he said. “Attribute-based trading works only when clean, bond-level data spans the universe and a single trade can draw at once on the security master, positions, restrictions, axes, pricing, and market intelligence,” Middleton said.
“Those systems predate any expectation that they would talk to one another, so most of a trader’s preparation still goes into assembling the picture by hand,” he added.
Before AI can automate more of the workflow, he added, “the infrastructure has to come first.”
Larry Tabb, Head of Market Structure Research at Bloomberg, offered a similar perspective: “The promise of Agentic is large, but the reality is uncertain. We have been down this road often starting even in the late 80s, but we have never seemingly gotten there. That said, we are certainly closer than we have ever been in the past.”
Where AI Is Likely to Help First
According to market participants, the first wave of adoption is expected to target the repetitive, information-heavy work that consumes much of the trading day.
Kinak expects AI to begin with workflow augmentation: “The first wave will likely be workflow augmentation rather than autonomous execution,” he said, pointing to trader commentary, broker color summaries and organizing market data as natural starting points.
Over the next three to five years, he expects delegation to expand into “pre-trade analytics, market-color synthesis, alerting, exception detection and decision support.”
Costabile sees similar opportunities beyond the trading desk: “There are massive tech gaps in the post-trade part of the workflow where a lot of human interaction still happens,” she said.
“Although we are beginning to see more predictive technology e.g., identify false positives in the settlement process there are still “bodies being thrown at the problem,” she said.
“AI will likely have the impact of one day changing the makeup of the back office from worker bees to more analytical decision-makers as agents can filter out and predict issues leaders need to focus on,” she added.
Middleton also believes information gathering will move first: “The work that moves first is the information-heavy part of the job,” he said, citing pricing collection, dealer rankings, liquidity discovery, routing smaller liquid trades within established thresholds, exception management and post-trade analysis.
But he stressed that automation depends entirely on reliable data: “None of this survives a stale price or an incomplete position feed.”
Tabb also pointed to operations as an early opportunity: “Even today there are still many manual operations and reconciliations that occur when there are new products and systems rolled out,” he said. AI could reconcile multiple streams of data while allowing humans to verify significant discrepancies before progressively increasing automation, he said.
Beyond operations, Tabb expects AI to play a growing role in investment research: “Smaller company research, or less frequently traded names will be covered by AI.”
AI will also synthesize research from multiple sources before eventually contributing to investment ideas, he said.
The Human Still Makes the Call
While AI is expected to take on more of the workflow, market participants drew a clear line around decisions involving judgment, accountability and client responsibility.
Kinak said human judgment remains essential “where context, risk, block liquidity, counterparty, client objectives and market nuance matter.”
Middleton identified responsibility for client capital as one of the clearest boundaries.
“The work that stays human is the judgment closest to the portfolio and the client,” he said, including defining target risk profiles, managing large or sensitive risk transfers, maintaining dealer relationships and interpreting changing market conditions.
“The desk spends less time executing and more on managing risk, handling the hard cases, and feeding better positions into the investment process.”
Costabile also expects evolution rather than revolution. “I do think many more people will be using it, but from a risk standpoint I don’t expect anything extreme like desks autonomously trading,” she said.
Governance Becomes More Important
As AI systems become capable of taking actions rather than simply producing analysis, governance and explainability become increasingly important. “The biggest risk is that agentic tools shift the risk profile from ‘answer risk’ to ‘action risk,'” Kinak said.
Institutional trading desks need to understand “what information was used, what assumptions were made, what was ignored, what confidence level applied, and where uncertainty remained.”
For Middleton, the greatest risk comes back to data quality: “Any system that can place a trade is only as reliable as the price it acts on and the position it believes it holds,” he said.
He also warned against blurred accountability: “Someone approved the policy, someone built the tool, someone bought the model, and no one owned the decision.”
Before firms deploy agentic AI, he said, they should determine “who owns the model, who owns the outcome, and who has the authority to switch the system off.”
Costabile believes governance frameworks are still evolving: “AI is new-ish so a lot needs to be developed still at the industry and firm levels,” she said, while adding that explainability is improving.
“More explainability means more trust, adoption and use,” she said.
Tabb framed the risks in the context of previous market disruptions: “This is very experimental technology and it could go horribly wrong, especially in a period of turbulence,” he said.
“If models are trading your account, data goes astray… and the models don’t know the data is incorrect and continues to trade, sending orders to the market at wrong prices.”
An Evolving Trading Desk
While opinions differed on the pace of change, all experts described trading desks evolving rather than disappearing.
Kinak expects “fewer manual handoffs, more automated synthesis, more data-driven decision support, and more push-based intelligence delivered directly into the trader workflow…The desk of the future will still need experienced traders.”
Middleton believes automation will allow traders to cover more while focusing on higher-value work.
“The main gain is scalability, since automation extends how much each trader can cover and strengthens the capability of the desk’s human capital.”
He also believes firms will increasingly differentiate themselves through proprietary data. “As these systems grow cheap and common, holding them counts for little,” Middleton said. Competitive advantage will come from “a firm’s own execution history, positions, dealer relationships, and feel for how its desk trades.”
Costabile likewise expects steady, incremental change.
“Pieces of the workflow will become more automated where possible,” she said, while noting that “product and data maturity will continue to play a significant role in how quickly desks evolve.”
Tabb expects AI to reshape the industry without replacing the people at its center. “It will create winners and losers,” he said, while predicting continued consolidation across the investment industry.
Ultimately, he said, money management remains rooted in trust.
“At the end of the day, money management is about trust,” Tabb said. “And folks, no matter how much AI is used, they want to be comforted by humans, not robots. The idea that the investment advisor is going away is just wrong.”
