#TradeTech2026 moved the AI debate further into trading – beyond LLMs & SLMs into systems, structure, and people. The emerging edge is increasingly less about access to data and models but how effectively firms integrate modular, real-time infrastructure to embed intelligence into workflows faster and more efficiently.

As automation increases, the trajectory is also moving to more multi-asset, agent-influenced trading environments. This in turn is shifting how liquidity is provided and by whom – the recent numbers from Jane Street emphasises just how swiftly markets are switching to tech domination.  The real constraint is not AI availability, but the speed and quality of organisational adaptation – how quickly firms can re-platform legacy systems, operationalise new architectures, and upskill talent to operate in real-time, increasingly AI-driven markets. Here’s what I learnt last week at Tradetech Amsterdam 2026:

1. AI as the real-time decision layer – not just analytics

Across panels, the conversation moved from future potential to active deployment. AI is now embedded in workflows as adaptive, goal-driven systems that continuously test and refine decisions. One key use case was linking analyst historical research with real-time signals (earnings calls, sentiment, flow, liquidity) to validate whether an investment thesis still holds but AI can be implemented across the investment life cycle: from idea inception, to decision support (pre-trade TCA/allocations), execution optimization (signal/alert based, timing, human assisted, not fully automated) to post trade analytics and reconciliation.

What this means for trading:

• Shift to include real-time validation of investment decisions in execution performance enabling faster adaptation
• Traders move from order handling to interpreting flow, liquidity, and market context alongside AI becoming central in linking research, data, and execution – the new agentic eyes & ears of the market – and the strongest argument against outsourcing a trading desk.

2. The data edge shifts from access to understanding and control

From panel discussions it became clear that the industry does not lack access to data – but rather effective access, structuring, and interpretation. As firms move away from vendor feeds towards ever more granular datasets (from Level 1 to Level 3) and increasingly unstructured inputs, the bottleneck is then the ability to extract signal from noise. The solution – the ability to audit data, and treat this as a controlled, continuously evaluated system with lineage, auditability, and clear ROI-linked use cases. Read more here.

What this means for trading:

• This increases opportunity for continuous thesis testing and more evidence-based execution
• Ability for traders to build on structured outputs, from raw data but overlaid with internal bespoke information
• Firms that can industrialise data quality (internal and external) can build scalable control layers for AI-driven trading

3. Trading architecture is shifting to modular, event-driven systems

Legacy OMS/EMS platforms remain slow, costly, and inflexible – even minor changes require weeks and significant investment. The shift is towards modular, API-driven architectures evolving into fully event-driven, real-time systems where data is streamed, pre-processed, and continuously updated. Firms are beginning to rearchitect workflows starting with high-value use cases and scaling through structured integration, embedding AI rather than treating it as an overlay using interfaces integrated with prompt based LLMs to lower the friction on interacting with complex systems. Read more here.

As traditional architectures rely on tightly coupled systems, this can create inconsistencies and latency through repeated data reconciliation. Event-driven design reverses this: systems react to events (e.g. order updates), triggering pre-computed downstream actions such as compliance checks. Moving processing to centralised models, reduces latency and increases efficient throughput. Streaming frameworks allow services to pull only relevant data, enabling flexible recombination without duplication. This also enables incremental transformation – deploying modular services iteratively rather than committing to large-scale rebuilds. Failures are isolated at the service level, improving resilience, while integration with third parties is simplified via gateway layers.

What this means for trading:

• “Buy vs build” becomes “orchestrate and integrate” making firms less of third-party consumers and more integrators of vendor applications
• Interoperability, APIs, and standards such as FIX Protocol become strategic as an integration layer
• Incremental deployment can replace large-scale transformation risk improving operational resiliency

4. Multi-asset trading becomes the default operating model

As data and workflows standardise, the importance of traditional asset-class-specific trading strategies diminishes. Execution becomes cross-asset and workflow-driven, shifting from siloed design to modular, capability-based components that are less dependent on locating liquidity in a specific instrument or asset class creating more of a plug and play approach to executing an investment decision.

What this means for trading:

• Trading becomes more systematic and portable across asset classes where liquidity is accessed by workflow – not venue or asset
• Desks now need to operate as true multi-asset platforms, not silos
• Differentiation shifts to speed, adaptability, and signal integration

5. The constraint is then organisational: skills, governance, and mindset

Further AI adoption is inevitable but scaling it is constrained by organisational readiness – not technology. Governance, integration, and education are now critical. As highlighted in a presentation by @Hugh Spencer at Janus Henderson, meaningful transformation requires deliberate investment in people alongside systems – this is our “Moneyball” moment and there’s a danger we’re not Brad Pitt but the ageing scout resisting change – as an industry we need to adapt or die.

What this means for trading:

• The urgent need to upskill traders into hybrid profiles (data, technology and market structure)
• Firms must actively create time for learning – this will not happen organically
• Governance is central: explainability, control, and auditability of AI decisions
• Competitive advantage comes from combining technology, talent, and mindset – not just deploying AI and that needs to come from leadership

AI in trading is evolving from assistive tools to embedded, goal-driven systems that can plan, reason, and act. However, scaling is constrained less by model capability and more by integration, governance, and organisational readiness. The argument being that the firms that succeed will be those that can build production-grade, interoperable systems with governance at the core – then scale AI. This shift from look back and check to continuous management is also evident in the shift in regulatory direction as highlighted this week from both Europe and the UK with increasing emphasis on testing, reliability, governance, and control frameworks – Read more here.

Thanks again for reading – more to follow next week and as always, all feedback welcome.

Rebecca