Last week’s newsletter set out five structural shifts shaping the future of AI and markets in 2026 (https://www.mindfulmarkets.ai/ai-in-trading-2026-five-structural-shifts-to-watch/). This week, the focus is on what market participants are already starting to see in execution behaviour, governance conversations and organisational decision-making. The message is consistent: AI is not a future prospect: iIt is already embedded in market dynamics and ideas on the potential impacts need discussion now. While AI is clearly beginning to reshape interfaces, workflows, and decision-making, most organisations remain stuck: building AI systems is no longer the hard problem; aligning people, skills, governance, and organisational change is. Here’s what I learnt this week on AI in Trading:

1. AI in Trading Is Already Here & Shaping Market Behaviour

AI-assisted execution is moving from experimentation to standard practice on trading desks. At CQG, AI-driven algorithms embedded in the execution stack are reported to predict S&P 500 moves with roughly 80% accuracy and reduce slippage by around $21 per trade, illustrating how AI-generated signals are now being combined with traditional risk controls (https://www.linkedin.com/posts/jeremy-grant-6b0219_cqgs-alli-brennan-on-ai-driven-algos-prediction-activity-7414343631097970688-yctN?utm_source=share&utm_medium=member_ios&rcm=ACoAAARqE5IBeeTUq6qzU4Eqccq_UO6-OMeR6Ao). Beyond execution, CQG is applying AI across operations, client servicing, and sales analytics, while maintaining infrastructure capable of supporting both institutional and retail flow at scale: AI-driven signal generation is now embedded in core trading workflows, not operating at the margins. The implication is not simply faster trading, but structurally different market micro-dynamics as machine-generated views increasingly interact with one another.

2. 2026 – From Pilots to Production

If 2025 was the year of pilot programmes, 2026 is shaping up as the year of production. The industry is leaving the “magic” phase of AI and entering an industrial phase. Many of last year’s forecasts are now being validated with one example from Nomura underscoring how quickly the conversation has moved from experimentation to execution (https://bit.ly/4qkmX0W). The firm describes a decisive shift toward production-grade AI platforms designed for scale, mass adoption, and bank-wide transformation. Frontier model releases have moved internal debate away from “which model” toward “how do we extract value.” Senior management expectations have risen accordingly and tolerance for experimentation without measurable outcomes is diminishing, with clear emphasis on “change-the-bank” initiatives and demonstrable return on investment. Technical capability is no longer seen as the binding constraint; the limiting factor is organisational readiness – skills, literacy, and change management. Nomura’s response has been to formalise global AI teams, platform strategy, and budgets, alongside a bank-wide AI literacy programme positioned as a critical enabler for 2026–27.

3. Agent Adoption & the Ability to Reverse

As agents become more autonomous and more tightly integrated into operational workflows, a new risk is emerging: the erosion of institutional memory and the diminishing ability to reverse or unwind decisions. Highly capable agents can optimise processes to a point where human intervention becomes difficult, obscuring accountability and control. The core challenge for institutions is not whether to automate, but how to design decision layers that maintain real-time visibility of operational risk and clearly define when to pause, reroute, or stop execution. While efficiency gains are tangible, they must be matched by resilient operating frameworks that preserve human oversight and control. Recent work from the Cambridge Centre for Alternative Finance underscores this issue and invites participation in its latest survey here – https://www.jbs.cam.ac.uk/faculty-research/centres/alternative-finance/research/live-research-surveys/ai-in-financial-services-2030-global-surveys/

4. Lessons from Singapore: Scaling AI Responsibly

Singapore’s MAS consultation on AI Risk Management Guidelines, along with regional initiatives under APEC, and the establishment of Korea’s Asia-Pacific AI Center all point to a region combining rapid adoption with increasingly operational governance expectations. Gary Ang’s reflections on AI risk supervision capture a shift many institutions are now experiencing. He argues AI governance often begins as something abstract and aspirational (a teddy bear) but in practice, it increasingly resembles a toy robot: functional, sharp-edged, and grounded in real failure modes (read more here – https://www.linkedin.com/pulse/personal-reflection-ai-risk-management-gary-ang-phd-nyeoc/). His recent work frames AI risk management to three enduring questions familiar from financial risk discipline: what is at risk, how is it managed, and who is accountable. The parallels with the Global Financial Crisis are clear – the crisis did not create risk; it revealed exposures that organisations did not fully understand. The same applies to AI today. Inventories, proportional controls, continuous testing, and clear ownership are emerging not as regulatory box-ticking exercises, but the ability to scale responsibly.

5. Not just Regulation.

For AI to operate reliably in markets, market structure – not just regulation – must evolve. Progress toward a usable consolidated tape in Europe is underway, but robust AI also requires high-integrity timestamps, harmonised identifiers, and consistent order semantics. These are foundational for systems that can learn and be audited. Fragmentation does more than raise costs; it undermines learning by driving models to overfit venue-specific behaviour rather than durable signals. Safe deployment also depends on credible simulation and replay environments, clear governance over model changes, and transparency around implementation, including whether firms can export, inspect, and reason about their own organisational logic or are locked into vendors. Without open standards and end-to-end auditability, AI risks narrowing rather than expanding intelligent execution. Given the speed at which AI is developing in trading, please join us at the FIX AI Working Group to discuss how we can best make this happen.

As always, thank you for reading – and happy January.

Rebecca