AI is no longer sitting on top of trading systems. It is reshaping the full trade lifecycle, from execution venues and compute layers to liquidity formation and now settlement architecture – reframing execution strategy away from just the order book to the full trade lifecycle – and creating a structural redesign of markets in the process. Here’s what I learnt this week on AI in Trading:

1. On-chain public equity trading goes live: BitGo and Figure complete first regulated tokenised public equity trades

BitGo and Figure Technology Solutions announced the completion of the first regulated, blockchain-native public equity trades executed on Figure’s Alternative Trading System (ATS). The equities were issued, traded, and settled fully on-chain demonstrating how tokenized equities can operate in a continuous, on-chain environment without compromising on risk management or regulatory standards.

Why this matters for Trading: While this appears not immediately related to AI in Trading – it is an important evolution which is likely to accelerate adoption. When public equities can be traded and settled within a programmable environment, AI systems can move beyond optimising order routing alone. They can coordinate and optimise execution, custody, collateral management and lifecycle processing, ensuring execution strategies don’t just focus on trade matching, but can now dynamically incorporate margin timing, collateral allocation and capital efficiency in real time.

By collapsing layers of intermediated reconciliation in batch-based infrastructures, the BitGo–Figure model materially reduces operational friction and enhances capital efficiency. Post-trade functions transition from administrative overhead to an integrated component of execution strategy – giving AI-driven systems yet another structural advantage. Read more here:
https://www.stocktitan.net/news/BTGO/bit-go-and-figure-complete-first-tokenized-equity-trades-on-figure-s-74z7gi7ng7id.html

2. Exchanges are upgrading AI-native trading: SAAS to MAAS

London Stock Exchange Group (LSEG) this week launched Model-as-a-Service (MaaS), a platform designed to host, distribute, and govern analytical models within market infrastructure – read more here https://www.lseg.com/en/media-centre/press-releases/2026/lseg-launches-model-as-a-service-and-welcomes-societe-generale. The announcement comes amid some bold claims circulating on LinkedIn that AI can now replicate the work of institutional quant teams with 12 structured Claude prompts effectively replacing a $400,000-per-year quant researcher in minutes. The claim: individuals can now build “quant models hedge funds pay millions to develop” simply through prompt engineering. Read more here – https://www.linkedin.com/posts/suleiman-najim-87457a211_breaking-ai-can-now-build-ml-models-that-activity-7429293051501662208-Exmw/?utm_source=share&utm_medium=member_desktop&rcm=ACoAAARqE5IBeeTUq6qzU4Eqccq_UO6-OMeR6Ao

Why this matters for Trading: if advanced trading models become increasingly commoditized through AI, competitive differentiation shifts away from model creation and toward validation, integration, governance, and execution. LSEG’s MaaS signals that exchanges and market infrastructure providers are positioning themselves as trusted distribution layers for models, embedding them within regulated, secure environments rather than leaving them as ad-hoc outputs from chat interfaces. For secondary markets, this matters: model access may become democratized, but performance, accountability, infrastructure coordination, and regulatory traceability will determine who can deploy those models at scale but still retain oversight.

3. Innovation meets Reality: Risk & Cost

Before trading desks are relegated to the history books and replaced by autonomous trading bots, two key challenges remain: risk and cost.

The rising risks of AI in Trading is actively being examined within the FIX AI and Complex Algo discussions – see previous newsletters here https://www.mindfulmarkets.ai/latest-newsletter/ and also this linked article from Sam Livingstone here https://www.linkedin.com/posts/slivingstone_it-is-becoming-increasingly-clear-that-people-activity-7429442028666695680-rqoU/?utm_source=share&utm_medium=member_ios&rcm=ACoAAARqE5IBeeTUq6qzU4Eqccq_UO6-OMeR6Ao).  However, alongside risk, cost is becoming just as critical a factor – as it will determine who has access to markets. Despite heavy investment in AI infrastructure, high-performance compute pricing remains opaque and inconsistent. Firms can pay materially different prices for similar hardware according to this recent article – https://www.forbes.com/sites/kolawolesamueladebayo/2026/02/18/why-ai-compute-pricing-remains-opaque-despite-massive-investment/, without greater transparency, AI implementation risks becoming more complex and expensive than existing EMS infrastructure issues.

Why this matters for Trading: As AI increasingly becomes part of secondary markets – compute access – not just capital access – increasingly determines competitive positioning and who has access to liquidity provision. Retail-facing platforms such as Axis Quant AI offer alternative access models  using dynamic, API-integrated AI to analyze real-time data and execute trades in a closed loop (https://www.axisquantai.com/). All of this could have long term radical impacts on how market access is provided and sustained – and why connectivity to markets will become just as important as the trading models and underlying execution strategies. Watch out for more to follow on this in a separate forthcoming study from Market Structure Partners.

4. Orchestration overtakes Modelling

As knowledge spread as to the capabilities of AI in Trading – there is also a growing sense of realism. As per this LInkedin Article “C𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲 𝗶𝘀𝗻’𝘁 𝗺𝗮𝗴𝗶𝗰” – https://www.linkedin.com/posts/ramanshrivastava_%F0%9D%97%96%F0%9D%97%B9%F0%9D%97%AE%F0%9D%98%82%F0%9D%97%B1%F0%9D%97%B2-%F0%9D%97%96%F0%9D%97%BC%F0%9D%97%B1%F0%9D%97%B2-%F0%9D%97%B6%F0%9D%98%80%F0%9D%97%BB%F0%9D%98%81-%F0%9D%97%BA%F0%9D%97%AE%F0%9D%97%B4%F0%9D%97%B6%F0%9D%97%B0-activity-7428145301179392000-3A8S/?utm_source=share&utm_medium=member_ios&rcm=ACoAAARqE5IBeeTUq6qzU4Eqccq_UO6-OMeR6Ao the argument is that we need to see advanced agents as structured decision loops rather than autonomous intelligence. Claude functions as a reasoning engine that decides which tool to call; external code executes the action and feeds the result back into the loop. Performance depends on tool integration, context management, and execution discipline. The architecture is domain-agnostic – only the tools change. The competitive edge therefore lies in orchestration (calling the right tools at the right time) and minimizing latency variance or jitter, not simply selecting the largest model.

NVIDIA’s research argues that most agentic workflows are narrow and procedural. Tasks such as formatting, validation, or API calls do not require frontier-scale models. Small Language Models (under 10B parameters) often deliver superior latency, cost efficiency (10–30x cheaper per token), and controllability – read more here https://arxiv.org/pdf/2506.02153.

The second paper – ToolOrchestra demonstrates this in practice. An 8B “router” model dynamically allocates tasks between lightweight tools and larger reasoning models. In benchmark testing, this orchestrated system outperformed a monolithic GPT-5 baseline while being approximately 2.5x more efficient and operating at roughly 30% of the cost. The implication is structural: scalable AI systems are built through intelligent routing, coordination, and compute discipline – not model size. Read more here – https://arxiv.org/pdf/2511.21689.

Why This Matters for Secondary Markets:  Trading is inherently latency and cost sensitive. As AI agents become embedded, competitive advantage will hinge less on access to the most powerful model and more on how effectively firms coordinate compute, data feeds, APIs, and execution venues. Deterministic orchestration, tight latency control, jitter minimization, and disciplined allocation of compute across workflows will define performance. Poor orchestration raises both infrastructure costs and market risk, while tightly engineered, low-latency systems create structural advantages.

5. Why Agents Maybe Changing Everything – but we still need to get the basics right first

Before AI can transform secondary markets, the fundamentals must be rebuilt and strengthened. Execution infrastructure matters. AI may provide analytical capability, but financial markets run on good data, FIX protocol adherence, coordinated multi-party workflows, curated reference data, and trust – none of which can be replaced by generating code in a chat window. SaaS models are not disappearing overnight, and operational infrastructure remains the backbone of market function.

What is changing is where AI is embedding. Agents are moving beyond analytics into core market infrastructure – AI-driven matching engines, automated compliance layers, and programmable liquidity frameworks. Secondary markets are evolving into engineered liquidity environments where compute capacity becomes a systemic input and settlement design increasingly converges with execution logic. Competitive differentiation will come from infrastructure capability, not just model deployment.

This shift also reframes risk. The focus moves from model accuracy to system architecture, resilience, concentration risk, controllability, and critically – accountability. The latest from FINRA makes it clear that AI-driven decisions must be explainable, reviewable, and attributable to a responsible human. Audit trails, ownership, and transparency are essential. Guardrails are not governance; traceability is – read more here https://www.finra.org/rules-guidance/guidance/reports/2026-finra-annual-regulatory-oversight-report.

In AI-enabled secondary markets, infrastructure design becomes strategy. There is no shortcut. The underlying plumbing – connectivity, execution pathways, controls- need to be carefully rebuilt. The technology is advancing rapidly, but durable transformation will depend on getting the basics right first.

As always – thank you for reading – any comments/feedback please let me know – all welcome!

Rebecca