Something for the Weekend: Trading Rewired – From AI to Quantum and the Rise of Intelligent Market Agents

Conversations in London last week with traders on how AI will be implemented in trading reinforced just how much confusion still exists – not just around AI’s potential, but the practicalities of deploying it beyond using co-pilot tools.

The next wave of AI in trading isn’t about making equity execution faster or smarter – it’s about taking electronic trading beyond equities, liquid bonds, and listed derivatives into markets that have traditionally been illiquid and voice-driven, such as commodities, private credit, and structured assets.

With natural-language interfaces, richer contextual data, automated risk analytics and built-in compliance, traders can draw on a wider set of insights across asset classes and decide when and where to trade more intelligently. This evolution expands the boundaries of what can be traded electronically, allowing liquidity to form dynamically and reshaping how price discovery, risk management, and market participation operate – while raising new challenges for governance, transparency, and regulation. Here are the 5 things to consider this week on the impact of this transformation:

1. AI to Quantum – Technology becomes Core

The era of “technology as a function” is over. For asset managers, data aggregation, validation, and analytics are no longer support the business – they are the business itself. Yet many firms still treat technology as an operational cost rather than a strategic engine of value creation. As highlighted in last week’s newsletter, the FCA’s new research programme into Quantum Computing in Financial Services marks an early signal that regulators are already thinking beyond AI, towards the computational architectures that will define trading in the 2030s. Their first discussion paper on the subject, released this week, sets the stage for how market participants might prepare for quantum-driven portfolio optimisation, pricing, and encryption (https://www.fca.org.uk/publications/research-notes/quantum-computing-applications-financial-services). Crucially, it calls for quantum readiness -coordinated action across firms, vendors, and regulators to ensure innovation is transparent, explainable, and resilient. Noting of course the industry’s pushback on the IBM claims highlighted here by @Nick Dunbar (https://www.globaltrading.net/experts-pour-cold-water-on-hsbc-quantum-computing-breakthrough/) the research suggestions the question is no longer if quantum will impact financial services, but how prepared firms will be when it does.

2. Cost, Context, and Co-Pilots – The initial pushback on AI adoption from trading desks has been a) the cost and b) what to implement to improve trading processes. @Perplexity may have just changed the game. The decision to make its Comet AI browser free could enable firms to use AI browsers to create context-aware research and execution assistants. Traders could use natural-language queries like “show me recent secondary transactions in AI infrastructure startups at >$2B valuation” to access synthesized insights from filings, LinkedIn hiring data, and SEC disclosures – streamlining trading ideas and due diligence without switching between platforms. As free AI interfaces like Comet evolve, could they finally disintermediate legacy data providers? @Ravenpack have also released their new research offering Earnings Intelligence Factors — RavenPack. Given the market structure focus on the potential for research rebundling, this was another interesting piece of news – @Substantive’s Research on the growth of private markets datafeeds and subsequent rising costs, up 40% (https://substantiveresearch.com/insights-and-press/private-markets-data-fees-soar-by-up-to-40-as-vendors-impose-take-it-or-leave-it-renewals-new-study-finds/) Clearly the demand for research and insights is only rising but what will firms choose to pay for in the future?

3. Data Deluge and Digital Fragility – how will this play out on trading desks? The #1 issue will be the ability to manage the explosion of trading data, exposing how fragile fast and automated markets can be when systems are flooded with corrupted or low quality information – read more from @Etienne Mercuriali on https://www.globaltrading.net/eurex-denies-getting-swamped-by-corrupted-messages-as-hft-dispute-escalates)  Unsupervised AI models could amplify these risks, creating self-reinforcing bias and an illusion of precision without genuine performance gains. A more effective approach will need to ensure human oversight with assistive intelligence – AI systems designed to support, not replace, trader judgment.

As trading speeds accelerate and market infrastructure faces mounting pressure from high-frequency volumes, regulators are stepping up with news that the FCA have requested two trading firms to supply formal “attestations” to address deficiencies in their algorithmic trading control frameworks FN London+2pub-ace-single-page-app.vir.onservo.com+2 . The European Securities and Markets Authority (ESMA) has announced that data infrastructure resilience will be a central pillar of its 2026 work programme (https://www.esma.europa.eu/press-news/esma-news/esma-2026-work-programme), with plans to deploy AI-driven supervisory tools for anomaly detection and market abuse monitoring. The move comes as markets transition to T+1 settlement and more centralized reporting, demanding that firms process and reconcile data faster and more securely. Under DORA, there will be stricter rules on ICT, cloud, and data service providers, tightening controls around outsourcing, vendor risk, and system reliability. One additional consideration for firms will be the impact from greater transparency and semi-public disclosure of trading and holdings.

4. McKinsey’s Six Infinity Stones of Agentic AI

For those building their own trading “superhero” agents – McKinsey’s latest paper introduces the Six Infinity Stones of Agentic AI – the foundational powers needed to deploy autonomous agents responsibly:

1. Prioritise workflow design over technology
2. Recognise agents aren’t always the right solution
3. Combat “AI slop” through evaluation and trust
4. Make every step trackable and verifiable
5. Reuse success systematically
6. Keep humans in the loop even as their roles evolve

As Stuart Winter-Tear noted in a recent LinkedIn post, much of this may sound like basic advice – “use the right tool for the right task” but most fact-checking frameworks assume a single, fixed truth, yet in emerging or novel markets, truth is often conditional. Small logical errors can compound, producing outsized and unpredictable outcomes – a reminder that we all still have a lot to learn.

5. Shifting Views on Regulation? The global debate over AI regulation reveals a growing divide between oversight and innovation and where responsibility lies. The United States recently rejected proposals for international AI oversight at the United Nations, arguing that global control could hinder technological progress (NBC News). Reinforcing this stance, former President Trump’s Executive Order — Removing Barriers to American Leadership in Artificial Intelligence promotes a “forward-leaning and pro-innovation” approach, revoking or revising existing AI-related policies, including President Biden’s 2023 Executive Order 14110 on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence.

In contrast, California Governor Gavin Newsom has introduced a new state AI law which targets companies developing advanced AI systems with annual revenues above $500 million, requiring them to disclose adherence to best safety practices; report safety incidents, protect whistleblowers; and support a state consortium for “safe, ethical, equitable, and sustainable” AI research.

These conflicting approaches mirror what Jon Danielsson calls “The Paradox of Perfect Supervision” . In the age of AI, the drive for ever-tighter regulation risks eroding individual and institutional accountability. As Jon Danielsson argues, when supervision becomes overly focused on measuring and controlling risk, it creates an illusion of safety – shifting responsibility from those who take risks to those who oversee them. This mindset encourages uniform behaviour and dependence on regulators, leaving the system more fragile when real shocks occur. True resilience will require a cultural shift in how we think about risk – from seeking perfect control to fostering personal responsibility, judgment, and diversity in decision-making.  But without more transparency AI incidents are often opaque or under-reported – perhaps now is the time for robust cross-jurisdiction frameworks for incident logging / replay, to avoid systemic risk be built unseen.

As we move toward conversational interfaces, autonomous agents, and quantum-scale computation, its clear that AI is no longer a “toolbox” for trading desks – it’s becoming the architecture for how markets will function.

As always, thank you for reading, and let me know what you found most useful, what you disagreed with, and what you would like to see more of next time.

Best wishes

Rebecca