AI Agents in Capital Markets: Real-Time Trading and Research Automation

In the first half of 2025, capital markets have seen a surge in autonomous AI agents – multi-step, LLM-powered “bot teams” that research, trade, model risk and rebalance portfolios without human prompts at each step. Financial institutions and fintech firms are rapidly moving beyond simple chatbots to deploy agentic AI platforms that coordinate tasks end-to-end. For example, Thomson Reuters in June debuted CoCounsel, an agentic workflow engine for tax, audit and accounting that pulls from 20+ billion documents and “understands the goal, breaks it into steps, takes action, and knows when to escalate to human input”pymnts.com. Similarly, Swiss core-banking vendor Temenos unveiled an AI “Co‑Pilot” and a financial crime-management agent in May to automate product launches and sanctions screening for global banksfintechfutures.comfintechfutures.com. Startups are also raising capital: New York’s AgentSmyth closed an $8.7 million seed round in June to scale a platform of five autonomous trading agents, now serving 48 institutions with up to $50 billion in AUMbusinesswire.combusinesswire.com. In short, the agentic era is arriving in finance: banks and asset managers are piloting AI “virtual analysts” for real‑time portfolio insights, and alliances are forming to build shared agentic infrastructure (e.g. ten Wall Street banks jointly funding Aiera, a research‑distribution AI platformai-street.co).

This rapid activity reflects the large market impact projected for agentic AI. Research firm Market.us forecasts the “agentic AI” market to balloon at ~44% annual growth, reaching roughly $197 billion by 2034linkedin.com. Embedding AI agents into trading and advisory workflows could materially boost efficiency and profitability (Bank of America estimates a 20–30% profit lift for banks by 2030cdotimes.com). However, experts warn this also multiplies new risks: autonomous agents may find creative but misaligned shortcuts (e.g. illegal data access or insider trading strategiesreuters.comreuters.com) if not properly governed. This report surveys March–June 2025 developments – from product launches and pilots to acquisitions – showing how institutions are adopting AI agents for algorithmic trading, research automation, risk modeling, and portfolio optimization. We also highlight the underlying technical architectures (multi-agent LLM workflows, real-time data stacks) and the key industry players spearheading this transformation.

Background

“Agentic AI” refers to generative AI systems that act autonomously on goals, not just respond to prompts. Unlike conventional chatbots that output text, agentic AIs can ingest data, make decisions, and execute actions on behalf of users. A Reuters analysis defines AI agents as “goal-directed GenAI systems that act autonomously to perform tasks”reuters.com. In practice, financial AI agents might autonomously monitor markets, generate trading ideas, execute orders, or aggregate research across sources. They promise to combine the speed and scale of machine learning with domain expertise: for example, a team of LLM agents might debate a stock’s outlook, quantify risk exposures, and then place trades without human interventiontradingagents-ai.github.iotradingagents-ai.github.io.

Finance is a natural proving ground for agentic AI because of its data intensity and clear objectives. The wealth of structured and unstructured data – market feeds, filings, news, and social sentiment – can be processed by specialized AI analysts. In a TradingAgents research framework, for instance, separate LLM agents play roles such as fundamental analyst, sentiment analyst, technical analyst, researchers (bullish vs bearish) and traderstradingagents-ai.github.iotradingagents-ai.github.io. These agents “collaborate through structured communication and debates” to produce buy/sell decisions, with a risk team validating exposurestradingagents-ai.github.iotradingagents-ai.github.io. This mimics a human trading desk or portfolio team, but with LLMs doing the heavy lifting of data analysis. Even outside trading, AI agents are being built to automate compliance, legal reviews, and customer advice. For example, early examples include Google’s (released) Operator agent and Microsoft’s emerging multistep CoPilot platforms, which inspired finance-focused variants.

Economics also drives interest: AI agents can unlock efficiency and new products. Institutional investors are piloting bots that continuously scan global filings and news for portfolio signals, boosting research coverage. Retail platforms are testing LLM assistants that monitor portfolios and alert customers of opportunities. According to Thomson Reuters Ventures, “the use of AI to generate trade ideas, stay up to date with company coverage, and analyze investment data is no longer optional”businesswire.com. Market forecasts concur: global spending on autonomous AI agents is surging (see chart below), with projected CAGR ~30–40% over the next decadelinkedin.com【69†】.

*Figure: The global Agentic AI market is projected to grow rapidly. Analysts estimate agentic AI (including financial use‑cases) will reach tens of billions in annual revenue by the early 2030slinkedin.com (Source: Globe Market Insights chart).

As background, it is important to note that agentic systems also introduce compounded risks. Researchers warn that highly autonomous agents can “optimize results (e.g., stock trading) by engaging in illegal behavior” if not alignedreuters.com. Financial regulators and compliance teams will thus need new guardrails (see Strategic Outlook). For now, leading firms are imposing human‑in‑the‑loop gates, audit trails, and explainability in these systems to build trustpymnts.comfintechfutures.com. Overall, the last quarter has seen finance move from proof‑of‑concepts to concrete implementations of agentic AI, setting the stage for much wider adoption.

Recent Developments (Mar–Jun 2025)

• Thomson Reuters (June 2): Thomson Reuters launched CoCounsel, a new “agentic AI” platform for tax, audit and accounting professionalspymnts.com. CoCounsel embeds LLM agents into the workflow, linking internal knowledge bases, Checkpoint (tax code database) and IRS regulations into a single AI-enabled workspacethomsonreuters.compymnts.com. In pilots, firms report dramatic time savings: one client cut a multi-day residency and tax code review to under an hour by using CoCounsel to autonomously parse 36 states’ rulespymnts.com. Thomson Reuters emphasizes that CoCounsel agents “understand the goal, break it into steps, take action, and know when to escalate to human input”pymnts.com. The platform is powered by partnerships (OpenAI, Anthropic, Google Cloud) and TR’s 20 billion+ document basepymnts.com. TR plans to extend agentic versions of its legal/compliance products (e.g. Westlaw, Practical Law) later in 2025thomsonreuters.compymnts.com.

• KX (June 11): KX, the time-series analytics platform now independent from Refinitiv, announced the AI Banker Agent blueprint for global markets. Co-developed with NVIDIA, the blueprint packages NVIDIA’s NeMo LLM tools, Nemotron large‑model services and GPU-accelerated inference alongside KX’s high-speed time-series databasebusinesswire.com. It is designed for sell-side banks’ trading and advisory desks. Key features include an AI-Powered Research Assistant for on-demand report generation, an AI Relationship Manager for personalized client outreach, and a Personalized Portfolio Manager for dynamic asset allocationbusinesswire.com. NVIDIA highlights that this “easy button” agent harvests insights from vast tabular and unstructured data to advise bankers in real timebusinesswire.com. Importantly, the AI Banker Agent includes built-in compliance safeguards and human-in-loop controls, reflecting KX’s emphasis on regulatory rigorbusinesswire.combusinesswire.com. This announcement signals a push to embed AI agents into the front-office client process. (KX also launched an “AI Lab” at NVIDIA’s GTC Paris event to allow banks to experiment with these workflows.)

• Temenos (May 28): At its Temenos Community Forum in Madrid, Temenos unveiled two new AI products. The Temenos Product Manager Co-Pilot is a GenAI tool for product teams, integrated with Azure OpenAI, that can design and test new banking products. More notably, Temenos introduced the FCM (Financial Crime Management) AI Agentfintechfutures.com. This agent continuously screens transactions against sanctions and watchlists using an “agentic AI” process to reduce false positives. Temenos claims the FCM agent is “the first solution of its kind” to get regulatory approval for explainable reasoning, and it is already live at a “tier-one global bank”fintechfutures.com. Both tools were developed with customer partners (e.g. Commerce Bank, Varo Bank and BIL Luxembourg)fintechfutures.com. CEO Jean-Pierre Brulard stated Temenos will be “the first core banking company to massively invest in GenAI and agentic AI”fintechfutures.com, underscoring that multi-step AI agents are now a core part of their strategy.

• AgentSmyth (June 2): New York startup AgentSmyth raised a $8.7 million seed round (co-led by FinTech Collective and Thomson Reuters Ventures) to expand its autonomous trading agent platformbusinesswire.com. AgentSmyth’s system deploys a team of five AI agents running in parallel – each scanning different data sources (macro news, social sentiment, quant factors, options activity, earnings) – to generate actionable trade ideas for institutional clients. In under a year, AgentSmyth reported deployments at 48 institutions (banks, hedge funds, asset managers with $2–50 B AUM)businesswire.com. The platform continuously delivers “institutional-grade intelligence” by integrating real-time feeds into LLM analysis. CEO Pulkit Jaiswal emphasizes that by moving beyond static research, AgentSmyth’s agents directly tell traders how to trade on insightsbusinesswire.com. Notably, BNY Mellon’s Ascent innovation program has onboarded AgentSmyth to refine the platform for trading desks. This funding round signals significant venture confidence that multi-agent trading tools are ready for prime-time.

• FE fundinfo – Lunar AI (May 23): Financial data firm FE fundinfo (UK) acquired Lunar AI to boost its Nexus investment-management platformfintechfutures.com. Lunar AI (founded 2022) builds no-code AI tools for generating financial content via LLMs. FE fundinfo will integrate Lunar’s tech into Nexus (launched late 2024) to automate research content creation and data analysis for asset managers. The deal (terms undisclosed) is FE fundinfo’s fourth acquisition in a year to expand its AI suite.

• OakNorth Bank – OpenAI partnership (June 10): UK business lender OakNorth announced a partnership with OpenAI to develop “custom-built GPT tools” for broad deployment across its operationsfintechfutures.com. The bank stated these tools will enhance speed, personalization and efficiency in areas from digital lending to credit risk scoring. OakNorth’s sister firm ONCI has already used ML in its lending platform; this deal signals a jump into full LLM usage.

• Lloyds Banking Group (June 13): Lloyds appointed Aritra Chakravarty as Head of Agentic AIfintechfutures.com. Chakravarty’s mandate is to oversee the bank’s plan to launch a prototype agentic AI system “later this year”fintechfutures.comfintechfutures.com. Lloyds had disclosed in April that it is building its own generative AI platform on Google Cloud Vertex to support such agentic workflows. This hire – under new Director of AI Rohit Dhawan – shows a major UK retail bank moving from research to execution of multi-agent systems.

• Westpac (June 13): Australia’s Westpac named Dr. Andrew McMullan as Chief Data, Digital and AI Officer, effective Sept 2025fintechfutures.com. McMullan joins from Commonwealth Bank and will oversee AI initiatives, highlighting a regional C-suite focus on AI/agents (Westpac already runs generative AI pilots internally).

• FAYBL (May 28): Singapore-based FAYBL launched an autonomous General AI Agent for financial advisers, integrated with Iress’s Xplan wealth platformnasdaq.comnasdaq.com. This SaaS agent can generate tailored investment reports, answer client queries, and draft advice documents by pulling in client data and market insights. The product demo suggests advisors can type natural-language requests (e.g. portfolio review) and the agent produces complete narratives and trade recommendations. FAYBL’s offering shows agentic AI moving into wealthTech use-cases, augmenting advisors rather than robo-advisors.

• BlackRock Asimov (June 12): At its June investor day, asset manager BlackRock disclosed an AI research system nicknamed Asimov. Asimov is described as a “virtual investment analyst” that scans research notes, regulatory filings and internal documents overnight to generate portfolio insightsinvesting.com. COO Rob Goldstein said the goal is to “scale our people” by having these AI “agents” working while humans sleep. BlackRock plans to expand Asimov’s use firm-wide by their 2027 investor day, illustrating how leading asset managers are adopting agentic AI for fundamental equity research.

• Other Notable Moves: In May, ten major Wall Street banks co‑funded Aiera (a NYC fintech) with $25 M to build a shared AI research platformai-street.co. Aiera aims to let banks share research analysis via AI, solving “screen space” fragmentationai-street.co. In Asia, Ant Group announced a time-series transformer that can cut FX hedging costs by ~60%ai-street.co, a sign that transformer-based agents are tackling numerical finance problems (beyond language). And Moody’s Analytics told Bank Automation News (June 11) it is embedding agentic AI into its risk products to speed credit analysis and regulatory reporting; the agentic AI market itself is now pegged at ~$196 B by 2034linkedin.com. Together, these moves show a broad industry shift: from big‑tech AI partnerships (OpenAI, NVIDIA) to fintech incubators and traditional vendors, all racing to automate capital market functions with intelligent agents.

Technical Innovations

Finance-grade AI agent systems combine several emerging technologies and architectures:

• Multi-Agent Orchestration: Modern frameworks break down trading and analysis into teams of specialized agentstradingagents-ai.github.io. For example, in the TradingAgents model researchers at UCLA conceptualized groups of LLMs with distinct roles – e.g. “Fundamental Analyst”, “Sentiment Analyst”, “News Analyst”, “Technical Analyst”, plus Bull/Bear Research teams – each working on parts of the market puzzletradingagents-ai.github.iotradingagents-ai.github.io. These agents communicate via structured protocols: analysts compile reports, researcher agents debate bullish vs bearish cases, trader agents make buy/sell decisions, and risk-agent teams vet any trade against limitstradingagents-ai.github.iotradingagents-ai.github.io. In practice, real systems use similar patterns. Figure 1 below (from the TradingAgents project) illustrates a hypothetical multi-agent trading workflow: data feeds into analyst agents, flows through researcher debates, and ends with trader execution and risk assessment.

  Figure 1: A multi-agent trading architecture (TradingAgents framework). LLM-based agents specialize as Analysts, Researchers, Traders and Risk Managers. They interact via structured outputs and debates to produce tradestradingagents-ai.github.iotradingagents-ai.github.io.

• Prompting & Model Selection: Key to these systems is how agents generate and share information. Rather than pure free-text chat, many designs use structured “report” formats and chain-of-thought debate. For example, the TradingAgents framework’s agents produce concise analysis reports and reason in a ReAct-style looptradingagents-ai.github.io. Designers choose lightweight LLMs for data retrieval (fast, cost-effective) and deeper LLMs for complex reasoning. TradingAgents notes that quick “retrieval” models handle price feeds and news lookup, whereas stronger “reasoning” models deliberate strategy, allowing the system to avoid constant GPU compute while maintaining accuracytradingagents-ai.github.io.

• Real-Time Data Integration: Financial agents must ingest live feeds (market prices, news alerts, social sentiment). Technology stacks now include high-performance time-series databases and streaming AI inference. For instance, the KX AI Banker Agent blueprint uses KX’s millisecond latencies and NVIDIA’s accelerated LLM inference (NeMo/Nemotron) to ensure agents can act on up-to-date databusinesswire.com. Similarly, Temenos’ FCM AI Agent runs continuously against live transaction streams, scoring each new trade against watchlists in real timefintechfutures.com. These workflows often rely on vector databases or optimized retrieval layers to ground LLMs in current context.

• Domain-Specific Tooling: Many agents integrate specialized knowledge sources. Thomson Reuters CoCounsel connects its LLM agents to proprietary legal and tax databases (e.g. Checkpoint code, IRS guidelines)thomsonreuters.compymnts.com. Agents may also call external APIs: e.g., a trading agent could trigger order-execution APIs or risk models automatically. The KX/NVIDIA stack includes compliance and audit modules, and Capgemini is working to codify best practices for agentic AI in banking.

• Explainability and Oversight: Given the stakes, explainable AI is a priority. Thomson Reuters and Temenos both highlight that their agents produce audit trails and human-readable reasoning. Temenos’ compliance agent provides “explainable intelligence” with probability scores to justify each alertfintechfutures.com. Thomson Reuters’ CoCounsel keeps full data lineage and integrates human-in-loop checkspymnts.com. Some systems offer “ghost mode” testing (as Temenos does), running the agent in parallel to human analysts to build confidence before live deploymentfintechfutures.com. In essence, even as agents automate, banks are embedding human oversight at decision nodes to prevent misalignment or errors.

• Infrastructure and Compute: These advanced agents often require heavy compute. NVIDIA’s involvement is notable: KX agents run on NVIDIA’s NIM microservices and GPUsbusinesswire.com, and markets are training larger, finance-tuned LLMs for this purpose. Cloud platforms (Azure, Google Cloud) now offer agent-optimized services (e.g. Google Vertex AI) which banks like Lloyds are already leveragingfintechfutures.com. Firms also experiment with more efficient models: Ant Group’s time-series transformer (focused purely on numerical prediction) shows a parallel effort to apply AI cheaply in financeai-street.co.

In summary, today’s agentic trading systems are multi-layered AI stacks: they combine LLMs with specialized modules (quant analytics, compliance checkers, execution engines), all orchestrated by a central agent logic. This modular architecture allows, for example, a Risk Guardian agent to intercept trades flagged by a Researcher agent if volatility spikes, mirroring human desks. The result is a highly integrated, real‑time AI “war room” for markets – as drawn in Figure 1.

Key Players and Ecosystem

The agentic AI wave in finance spans incumbents, fintechs, cloud providers and hardware vendors:

• Financial Data & Software Leaders: Thomson Reuters, Bloomberg, Refinitiv/TK (part of LSEG) and similar incumbents are race to embed agents in their platforms. Thomson Reuters (via CoCounsel) is the earliest mover with a live agentic product for accounting/compliancepymnts.compymnts.com. Bloomberg is reportedly exploring similar multi-step assistants (though specific launches in Q2 2025 are not yet public). Core banking and treasury software firms are also active: Temenos (as noted) and Finastra have pledged major GenAI investments, with Temenos already fielding agentsfintechfutures.com. Core platform vendors like Iress (via FAYBL integration) and SAP/Oracle are likely building agent features for wealth and corporate banking.

• Fintech Startups: A surge of startups is designing specialized agents: AgentSmyth (trading bots), Lunar AI (investment research templates), FAYBL (wealth advisors), Aiera (research distribution), and many others. These ventures often raise venture and strategic capital. For example, FinTech Collective and Thomson Reuters co-led AgentSmyth’s seedbusinesswire.com, and Microsoft/Third Bridge joined a Series B for Aiera backed by ten banksai-street.co. Fintechs are also incubated via accelerator programs (e.g. BNY’s Ascent for trading agents). Their agility is pushing new use-cases (e.g. Ant Group’s treasury AI, or startups using LLMs for portfolio optimization).

• Banks and Asset Managers: Large institutions themselves are deploying and co-developing agents. Big banks such as JP Morgan and HSBC are known to have internal AI labs (e.g. JP Morgan’s “LOXM” and “Ask D.A.V.I.D.” models for equities) and are now extending them into agentic forms. Lloyds and Westpac’s new AI leadership rolesfintechfutures.comfintechfutures.com show banks planning dedicated agent strategy teams. Asset managers like BlackRock (Asimov) and Goldman (reportedly using ChatGPT internally) are embedding agents into investment workflowsinvesting.com. Even custodians and infrastructure providers (e.g. DTCC, exchanges) are quietly testing AI for surveillance and risk.

• Cloud & AI Vendors: Companies like OpenAI, Anthropic, Google, Microsoft and AWS supply the foundational AI models. OpenAI’s GPT models already power parts of CoCounselthomsonreuters.com, and many banks rely on Azure/OpenAI or Google Cloud’s Vertex for hosting agent workloadsfintechfutures.compymnts.com. NVIDIA stands out by offering end-to-end agentic solutions: its GPU hardware, LLM frameworks (NeMo) and AI supercomputers are embedded in projects like KX’s and Temenos’. NVIDIA’s finance VP calls the KX agent “an ‘easy button’ giving relationship managers a competitive edge”businesswire.com. On the data side, firms like Palantir, Snowflake and KX/OneTick are enabling agents with rich live data access.

• Consulting and Systems Integrators: Major consultancies (Accenture, Capgemini, Deloitte, EY) are rapidly building AI trading/generative AI practice groups. Capgemini, for instance, has been named lead SI for implementing KX’s agent blueprintbusinesswire.com. These firms help banks customize and govern agent deployments, blending AI with legacy banking systems.

• Regulators and Alliances: Public and quasi-public players are engaging too. In the UK, the Financial Conduct Authority is sponsoring AI labs (with Nvidia) for financial firms. The European Securities and Markets Authority (ESMA) released a generative AI survey in April 2025, finding 85% of banks using tools like ChatGPT internallyai-street.co, prompting calls for clear AI usage policies. Industry consortia are forming: besides Aiera, bodies like the BIS Innovation Hub are exploring AI agents for fraud detection and resilience.

Strategic Outlook

Adoption Trajectory: Agentic AI is set to proliferate across capital markets through 2025–26. With dozens of pilots underway, many banks will move from experimentation to selective deployment. Early use-cases are clear: front-office research assistants, compliance monitors, and client-facing advisory bots. As generative AI becomes table stakes, agentic features will differentiate leading platforms. By next year, it is likely that most investment banks and asset managers will have at least a prototype agent operating under supervision. Global surveys and analyst reports predict that within 2–3 years, agent-driven automation will be common in trading floors and risk desksai-street.cocdotimes.com.

Competitive Impact: Firms that effectively harness AI agents could gain a sizable edge. Studies (e.g. by bank analysts) estimate 20–30% profit uplifts from AI agents by 2030cdotimes.com, through cost savings and new revenue (e.g. personalized advisory fees). For instance, AI-assisted research can cover far more securities with fewer analysts, and automated risk models can tighten margins by freeing capital. In wealth management, advisors augmented by AI agents could manage much larger client loads. Conversely, laggards risk falling behind as “AI-native” hedge funds and fintechs capture alpha fasterbusinesswire.com.

Risk and Governance: This upside comes with new hazards. Autonomous agents could inadvertently abuse privileged data or pursue extreme strategies. The Reuters legal analysis highlights potential for “misaligned” agent behavior: from privacy violations to hacking and even insider trading, if agents optimize without proper constraintsreuters.com. Therefore, internal governance is critical. Leading deployers insist on human veto points, explainable outputs and robust testing. For example, Temenos’s “ghost mode” lets banks compare agent decisions to humans until confidence is builtfintechfutures.com. Similarly, KX’s blueprint emphasizes compliance-ready safeguards built into every agent workflowbusinesswire.com. We expect institutions to codify agent approval processes, model risk policies and oversight committees in the coming year.

Regulatory Landscape: Regulators globally are still catching up. No specific AI-agent rules exist yet, but incumbents on regulatory bodies are issuing guidance. For example, ESMA is advising firms to train staff on AI principles, even as they use these tools internallyai-street.co. U.S. and UK authorities are observing how AI affects market manipulation and privacy. Industry groups (Wolfsberg Group, FMLC in UK, etc.) are drafting best practices for generative AI in finance. Banks will need to document agent systems rigorously to meet existing regulations (AML, fiduciary duty, data use laws). Those that proactively engage regulators and highlight explainability and human oversight will mitigate risk of sanctions.

Market Evolution: Looking forward beyond 2025, we foresee several trends:

• Vertical Agents: Firms are likely to develop industry- or function-specific agent networks (e.g. credit underwriting agents, compliance agents, market-making bots) rather than one-size-fits-all models. For example, an agent trained on fixed-income analytics may serve bond desks.

• Alliance Platforms: The Aiera example suggests banks may pool resources for common AI infrastructure (data consortiums for training or shared research). Such alliances can lower costs and create standards for things like citation and IP in AI outputs.

• Shifts in Talent: The human workforce will evolve. As Nvidia’s CEO expects, many routine analyst tasks will become baseline work for agents, and professionals must become “AI-Auditors” and strategistscdotimes.com. We already see banks hiring “Head of AI Agents” (Lloyds) and retraining staff to manage AI tools. Operational roles may shrink (e.g. manual reconciliation), but new jobs in AI ops, prompt engineering, and ethics are emerging.

• Technical Progress: The technology itself will advance rapidly. Ongoing research into LLMs for time series (e.g. Ant’s workai-street.co) and the creation of specialized “finance LLMs” will improve agent performance. Lower-latency inference (5G/edge cloud) will enable more real-time agent tasks. Continual learning pipelines may allow agents to update themselves from new data with minimal human retraining. We may even see cross-agent markets, where one firm’s risk agent service is consumed by another firm’s trading agent.

Conclusion: In summary, the Mar–Jun 2025 period has marked a turning point: agentic AI has moved from hype to reality in finance. We are witnessing concrete product launches, pilot deployments and strategic investments that embed LLM-based agents into the workflows of traders, analysts, and advisors. These innovations promise major efficiency and revenue gains, but must be managed carefully. Financial institutions that invest now in robust agent architectures, clear governance, and integration of these agents will likely set the pace. As one expert put it, “the age of applied AI in capital markets – this is no longer theoretical”businesswire.com. The coming months will show which approaches deliver real-world value and shape the new, highly automated landscape of capital markets.

Sources: Industry press releases and news (Mar–Jun 2025) on AI agent deployments in financebusinesswire.combusinesswire.comfintechfutures.compymnts.comnasdaq.cominvesting.comai-street.co, analyst and legal commentaryreuters.comtradingagents-ai.github.iocdotimes.com, and market research forecastslinkedin.comrobeco.com. Each cited excerpt is from March–June 2025.