Leading fixed income desks are embracing artificial intelligence and automation, but panellists at a recent industry summit warned that client obligations must remain the governing constraint as trading desks seek to scale.
The question of whether credit markets are structurally ready for the next wave of technological innovation dominated a buy-side panel at the Fixed Income Leaders’ Summit in Boston, with senior traders from Wellington Management, Legal & General Asset Management, T. Rowe Price, Invesco and JP Morgan Wealth Management offering a frank assessment of where the industry stands — and where it still needs to travel.
Jason Quinn chief product officer and head of sales at Trumid, who chaired the session, framed the discussion around four competing forces: artificial intelligence, the digitisation of block trades, automation, and the quality of data and pricing analytics. An audience poll placed AI at the top of the agenda, and it consumed the opening exchanges.
Masaya Okoshi, senior managing director and fixed income trader at Wellington Management with 13 years at the firm, drew a clear conceptual distinction between AI and the rest of the electronic trading toolkit.
“AI is learning about the unknown and incredible ability to leverage knowledge to help solve the unknown,” he said. “All this other stuff I think is like efficiency, productivity, and leveraging the knowns.”
But Okoshi was quick to anchor the discussion in the realities of institutional asset management. “At Wellington, and my colleagues here, we are fiduciaries of our clients’ [interests], so we need to ensure that our client dollars are in good hands, are secure, our firm is actively focused on supervision, governance, oversight — all the other factors that go into leveraging the powers of AI in a smart, safe way.”
Brian Rubin, head of US fixed income trading at T. Rowe Price said the firm’s posture was similarly one of active encouragement rather than restriction.
“We’re encouraging everyone to dig in, get involved,” he says. “Everyone’s on equal footing, you could create things, make things more efficient, not just relying on traditional technology, and it’s really a way that everyone has worth.”
Ryan Raymond, head of US trading at Legal & General Asset Management described a more granular rollout at his firm.
“We started, I guess, over a year ago with Copilot, and that probably similarly to many of you, now we all have Claude code on our computers. Everybody in trading, everybody in the front office has access to it to be able to build apps and dashboards.”
He explained that L&G is now trying to consolidate the outputs of that exploratory phase.
“If you’ve got a great idea, and I’ve got a great idea, and it overlaps by 50% – let’s build one cohesive unit,” he said.
Raymond pressed beyond efficiency gains to ask a more fundamental question: “So that’s great, and helps a lot on your day to day, but it frees up time. What do you do with that time? Are you talking to traders? What are you doing to help alpha generation? That’s really what it’s about.”
Quinn raised what he called the “big risk” of large language models (LLMs) their inherently non-deterministic nature and asked whether panellists could ever foresee an AI making autonomous investment decisions.
Okoshi was unequivocal on the principle, if nuanced on the trajectory.
“At the end of the day, our clients trust us with their capital. If have a very rare date with my wife, I might use Gemini to find a good restaurant and movie or something, but with my client dollars, I can’t look them in the eye and tell them Chat GPT made the decision of how to trade.”
He was also candid about the likely structural consequences for the profession.
“I think similar to how electronification disintermediated some average or lower average trading talent, I think similar levels of dispersion will happen within the more skilled, higher value-add [activities]. Winners will continue to differentiate, and a lot of the lower hanging fruit will ultimately get automated away.”
Rob Simnick, a senior credit trader at Invesco, said the portfolio trade had in practice become the primary vehicle through which block risk was being digitised. but identified a structural gap in how dealers rewarded information-sharing.
“All my career I’ve been told, ‘If you show us this block, just to us, we’re giving a better price to you’. I want to see us get rewarded for that.” He added that smarter, more targeted RFQ mechanisms where dealers are placed in narrower competition in exchange for better pricing commitment — could help bridge the divide.
Rubin backed the idea of decomposing large orders into smaller electronic executions, noting that technological constraints that once made this difficult were receding.
“If you have to be off the desk, you have to be in a meeting, you don’t miss a trade,” he said. “Automation is great – traders are supposed to be adding value, and the real value is not missing liquidity.”
Okoshi offered the most systemic reframing of the block debate. “I don’t really see a world of block trades and non-block trades, to be honest. During COVID, a US$5 million trade might have been a block trade, and now US$50 million may be a block trade. I look at all trades together in one group, identifying all the different factors that go into optimal execution outcomes, of which size is one — but we don’t think of the world [in those binary terms of block/non-block].”
Quinn turned to automation, observing that while rules-based workflow automation within RFQ protocols had become well-established, there was more to be done higher up the trade lifecycle.
Daniel Blonshteyn, managing director and US head of fixed income trading at JP Morgan Global Wealth Management described a desk that has already moved decisively into high-volume automation.
“Overall I think the journey started with automation where we’re basically now 80% automated — I think the right level is basically somewhere in the 80s, maybe mid-80s, where you want to be, and what that means is it’s zero touch, so a trade comes in, it gets executed, we set the rules beforehand,” he said.
The benefits, he added, had been transformational in scale terms.
“We’ve quadrupled our trade notional count, but our desks have shrunk by half, so we’re able to do a lot more with less, but at the same time we’re also able to provide service, and we’re not losing out on any commissions.”
He was careful, however, to demarcate the limits of the model.
“The goal is not to eliminate the trading desk, he said. “The trading desk is still there. There’s still situations that they have to work. They’re still responsible for every trade that gets executed — someone owns that trade, but they don’t have to touch it.”
Blonshteyn also identified the harder automation problem still ahead, extending no-touch execution into less liquid corners of the credit market. Okoshi described Wellington’s approach to automation using the language of risk governance rather than efficiency.
“We don’t call it no touch or automated, we call it more factor-based trading, if anything,” he said. “I trust the factors we apply to executing each trade a lot more than my own eyeballs.”
He used an analogy drawn from Tesla’s Full Self-Driving technology to illustrate why ecosystem-wide adoption matters as much as individual firm readiness.
“What scares me is not the technology in my own car being able to get from point A to point B, but it’s all the other terrible drivers out there – and you know what they can do to me and how my system will react to that,” he explained. “The more users use it, the safer it gets. That’s the model that we’ve taken with Trumid on FSD, and I’m very confident that this is only going to grow as more participants [join].”
Invesco’s Rob Simnick, outlined his vision for automation on a lean desk managing significant IG credit risk.
“There will be a point where a PM can send an order out based on some of the factors that the panel was talking about – whether it be liquidity or duration or size – those are going to go get filled, and the trader is going to be basically a monitoring system,” he said. “Less manual work, more oversight — which I think is the natural progression of us in this role.”
The panel’s final substantive topic — pricing analytics and data quality — drew broad consensus that the landscape had improved markedly, but that further gains were needed to support the next stage of automation.
“When I’m running pre-trade on a $100 million or a billion-dollar IG credit risk transfer, I can have this estimate of where I think it’s going to price, and especially on a year-to-date basis, it is correct to within a basis point, I think it’s incredible right now,” said Simnick.
He argued that dealers needed to be more willing to open up data flows to the buy side, and that pricing services needed to move beyond TRACE-sourced inputs.
“Trades that haven’t priced yet, CDX, other signals from other markets, these are all going to make these pricing systems better,” he noted. “That being said, I think we’re looking at incremental improvements as opposed to very transformational, because the pricing is predicting actual execution very well.”
Blonshteyn tied pricing accuracy directly to the viability of automation programmes.
“It’s very important,” he argued. “We couldn’t do it without being comfortable that the price is accurate. A large portion of the credit market is pricing pretty accurately on the pre-trade side, and a lot of the vendors that are here have done a good job of building tools around that.”
Raymond pointed to the real-time currency of dealer analytics as an underexploited resource.
“What I’d really like to see is the street use their analytics a little bit more, and be able to find ways to update those morning runs on positions, on their portfolio trades (PTs),” he said. “I see [dealers] doing PTs before they get their breakfast – that changes their position constantly, and I end up looking at old runs.”
As the panel closed, Quinn asked each participant what credit markets could learn from other asset classes. The answers converged on a single theme: the real opportunity lies not in further optimising execution, but in what trading professionals do with the time that automation frees up.
“As a fixed income trader at Wellington, the point of execution is less than 20% of my day,” saisd Okoshi. “Ten years ago, a portfolio trade used to take two weeks to complete, [moving] 500 bonds used to take three, four, or five hours to complete — all that stuff can now be done in 30 minutes. So, what are we doing with the other nine and a half hours of our day? Are we just riding the wave of efficiency, or are we going to identify new alpha opportunities and return opportunities for our clients?”
Blonshteyn, drawing on lessons from equity markets, urged fixed income professionals to think carefully about technology lock-in and legacy at a moment when the market is evolving quickly.
“Our equity teams learned the hard way — some technology sounded great seven years ago, and [seven years later] we’re still on a contract, so what do we do? I think being able to stay nimble in that space, because I think that we’re moving now a lot quicker than we were five years ago,” he said.
Rubin called for capital and focus to be directed at the market’s less liquid corners. “We need to get the less liquid asset classes, like securitised and bank debt, to function more like the other markets — and not be scared of it.”
Raymond was unambiguous about what the execution trader needs to become.
“I think you need to be an investor, you need to add value to your firm from an investment standpoint, first and primary,” he said. “If you’re not doing that, yes, you will be automated out of a job.”
©Markets Media Europe 2025
