There is a quiet contradiction inside modern investment management. Hedge funds and asset managers are among the most analytically sophisticated institutions in the world, yet many of them still operate through a software stack that looks less like a machine and more like an archaeological dig. Research lives in Bloomberg notes, Excel files, PDFs, Slack threads, email chains, Python notebooks, broker portals, data warehouses, OMS screens, risk reports, compliance archives, and PowerPoint decks. Every tool is defensible on its own. Together, they form a brittle operating environment that depends heavily on human memory, tacit coordination, and manual reconciliation.
That was already expensive before AI. It becomes a much larger problem when firms try to deploy AI agents.
The reason is simple. A human analyst can tolerate fragmentation because humans are good at stitching together context. A portfolio manager can remember that an analyst note in March contradicted a broker call in April, that the real reason for owning a stock was not the valuation model but a channel check, and that a factor exposure that looks excessive on paper was intentional because it hedged another part of the book. Humans can operate across messy systems because they carry context in their heads.
AI agents do not have that luxury. An agent needs explicit access to context, permissions, workflow state, memory, policies, system boundaries, audit trails, and human approval paths. If those primitives do not exist, the agent becomes either a glorified summarizer or an operational risk.
This is the issue investment firms are now beginning to confront. Generative AI adoption in hedge funds is no longer theoretical. According to AIMA and Marex research published in 2025, 95 percent of hedge fund managers surveyed reported using generative AI in their work, up from 86 percent in 2023. Meanwhile, 58 percent of fund managers expect generative AI to play a larger role in investment decision-making over the next year, up from just 20 percent in 2023. Perhaps most telling: 60 percent of institutional investors said they would be more likely to invest in a hedge fund that allocates a meaningful portion of its budget to AI research and implementation. The question has moved from whether investment firms will use AI to whether their operating systems can support it.
Most cannot, at least not yet. And the data suggests this is not unique to finance. McKinsey reports that more than 80 percent of companies investing in AI have not yet seen significant impact on their bottom line. The problem is rarely the model. It is almost always the operating environment around the model.
The Stack by Stage of Manager Maturity
To understand why, it is helpful to look at how the technology stack changes across the life of a hedge fund or investment manager.
At the pre-launch stage, the stack is improvised. A technical founder or aspiring portfolio manager may use Interactive Brokers, Python notebooks, spreadsheets, ChatGPT, a few data APIs, Notion, Google Drive, and Slack. The workflow may be clever, but it is rarely institutional. Research may be strong, but the operating environment is fragile. There is usually no clean lineage between an idea, the data that supported it, the portfolio decision that followed, the risk constraints applied, and the eventual performance attribution.
This stage is full of high-agency builders. Many can write code, scrape filings, build simple backtests, and experiment with agents. What they lack is not intelligence. What they lack is the machinery that turns an investment thesis into a repeatable operating process.
At the small emerging manager stage, roughly from tens of millions to perhaps low hundreds of millions of dollars in assets, the stack becomes more formal but not necessarily more integrated. Bloomberg or FactSet may enter the workflow. A fund administrator may be added. The manager may have an OMS, a prime broker, basic compliance support, and outsourced operations. Research may still happen in a mix of Excel, email, broker PDFs, notebooks, and shared folders. AI, if used, is typically layered on top as a personal productivity tool rather than embedded into the operating model.
This is where fragmentation becomes most painful. The firm is serious enough to require process, but too small to afford the internal engineering and operational teams that larger funds take for granted. The team may want to automate research, monitoring, reporting, and risk review, but every workflow crosses several systems. An AI agent cannot easily observe, reason, and act because the organization itself does not have a unified representation of its own state.
At the institutional emerging manager stage, perhaps from $100 million to $500 million in assets, the fund often has real operational maturity. It may use enterprise-grade service providers, more formal risk tools, outsourced middle office, professional investor reporting, and more robust data infrastructure. The fact that Clearwater Analytics acquired Enfusion for $1.5 billion in 2025 to create a front-to-back platform for hedge funds and asset managers is itself evidence that fragmentation is already a recognized, billion-dollar problem even before agentic AI enters the picture.
Yet even at this level, the AI layer remains poorly defined. A fund may have good accounting, good trading systems, and good investor reporting, but still lack an AI-native workflow layer that connects research, decision-making, portfolio construction, monitoring, approvals, and institutional memory. The systems may be integrated enough for humans to operate the fund, but not integrated enough for agents to operate safely within it.
At the mature hedge fund stage, above several hundred million or billions of dollars in assets, the problem changes again. These firms often have internal engineering teams, data platforms, risk systems, and compliance functions. They can build tools. They can hire AI talent. They can create internal copilots. But they also have organizational complexity. Multiple teams, regions, asset classes, PMs, analysts, risk managers, and operations staff produce a different kind of fragmentation: not just fragmented software, but fragmented institutional knowledge.
At this scale, the question is no longer whether the firm can afford AI tooling. The question is whether AI can be governed, trusted, audited, and integrated into existing investment workflows without increasing operational risk. BCG's 2026 Global Asset Management Report underscores the pressure: despite a near tripling of industry AUM since 2010, profit margins have remained flat at around 30 percent. Revenues have grown more slowly than costs, creating negative operating leverage. The traditional assumption that scale automatically improves profitability no longer holds. Firms need a fundamentally different operating model, not just more assets.
The largest multi-manager platforms offer a useful comparison. Firms like Millennium, Point72, Citadel, Balyasny, and others are, in a practical sense, portfolio manager operating systems. They provide capital, risk frameworks, execution infrastructure, operations, monitoring, data, and institutional discipline around PMs. Their success is not simply that they find talented investors. It is that they build machinery around talent.
That is the key insight. In investment management, alpha is only part of the problem. The operating system around alpha determines whether it can survive contact with capital, risk, markets, clients, and time.
The Hidden Cost of Human Coordination
A senior executive at a roughly $200 billion asset manager once told us that their organization estimated an average of about six headcounts for every client they onboarded. The number was not meant as a universal benchmark, but as a window into the hidden economics of institutional investment management. Every new client or mandate brings workflows: onboarding, data collection, legal review, reporting, customization, risk constraints, operational setup, client communication, monitoring, and ongoing service.
From the outside, investment management often looks like a business of ideas and performance. From the inside, it is also a business of logistics. BCG found that total costs across asset managers grew at a compound annual growth rate of 6 percent from 2022 to 2024, with the greatest proportion (30 to 40 percent of total costs) concentrated in investment management and trade execution. The increases came primarily from competition for talent and a shift toward more complex products.
This is where the old military line attributed to General Omar Bradley becomes relevant: amateurs talk tactics, professionals study logistics. In hedge funds, amateurs talk trades. Professionals study the operating machinery that allows trades to become portfolios, portfolios to become track records, and track records to become institutional businesses.
The same is true for AI. The amateur question is whether an agent can generate a good investment idea. The professional question is whether an agent can be safely embedded into an investment organization.
Can it access the right data without seeing the wrong data? Can it distinguish between a draft thesis and an approved investment view? Can it tell whether a risk limit was breached because of market movement, stale data, or intentional risk-taking? Can it explain what it did, why it did it, what sources it used, who approved it, and how its recommendation changed over time? Can another human or machine audit the entire chain?
Most current hedge fund stacks do not answer these questions well because they were not designed for them.
What AI-Native Actually Means
The term "AI-native" is often used lazily. It does not mean sprinkling ChatGPT into existing workflows. It does not mean replacing analysts with autonomous bots. It does not mean allowing software to trade without human oversight. In institutional finance, that is not merely unrealistic; it is dangerous.
An AI-native firm is better understood as an organization whose workflows, data architecture, permissions, memory, governance, and operating model are designed from the beginning for humans and AI agents to work together.
This distinction matters. AI-native does not mean fully autonomous. It means agent-ready.
A fully autonomous fund would imply that agents independently generate ideas, validate them, size positions, execute trades, monitor risk, respond to market events, and communicate with investors without meaningful human intervention. That is not a credible near-term model for most institutional managers. It ignores the reality of fiduciary responsibility, model failure, hallucination, regime shifts, operational errors, compliance constraints, and allocator trust.
An AI-native firm instead asks a different question: which parts of the investment workflow can be delegated, accelerated, monitored, or augmented by agents, and what controls are required for that delegation to be safe?
In such a firm, an agent might monitor earnings transcripts against a portfolio's thesis base. Another agent might track changes in factor exposure. Another might prepare a draft investment memo. Another might reconcile new filings against prior assumptions. Another might flag that a research conclusion is based on stale data. Another might prepare an allocator-ready summary of how a position thesis evolved through a drawdown.
Humans remain responsible for judgment, risk appetite, governance, and accountability. Agents expand the span of control.
This is consistent with how broader enterprise thinking about agentic AI is evolving. McKinsey describes the emergence of the "agentic organization" as a new operating model in which humans work together with virtual and physical AI agents, structured around end-to-end outcomes with embedded governance and human accountability. Their research emphasizes that organizations must rethink workflows, roles, governance, and data foundations rather than merely adopting individual AI tools. The problem, as McKinsey Senior Partner Alexis Krivkovich puts it, is that leaders "are trying to create agency in systems explicitly designed for human control."
That is the shift investment firms must internalize. AI-native is not a model upgrade. It is an operating model redesign.
Why Existing Processes Must Be Re-Engineered
The temptation inside many investment firms is to insert AI into the current process. Let analysts use LLMs to summarize filings. Let investor relations teams use AI to draft DDQ responses. Let PMs ask a chatbot questions about positions. Let developers use copilots to write scripts. These use cases are useful, but they are not transformative.
They are single-player AI.
The deeper opportunity is multi-player AI: agents embedded inside shared institutional workflows. That requires re-engineering the process itself.
Consider the research lifecycle. In a traditional discretionary equity fund, an analyst may identify an idea, build a model, read filings, consult broker research, speak with experts, write a memo, discuss with a PM, and track the position over time. The artifacts of that process often live across documents, chat threads, spreadsheets, meeting notes, and memory. AI can help summarize each artifact, but it cannot fully understand the process unless the process is represented in a machine-readable way.
An AI-native research process would capture the thesis, supporting evidence, counterarguments, assumptions, data sources, decision owner, approval state, portfolio exposure, risk constraints, and subsequent updates as structured workflow objects. That does not remove the analyst. It gives the analyst and PM a living operating record.
Now consider risk. In many firms, risk is both a system and a conversation. A factor report may show an exposure. A PM may explain why it is intentional. A risk manager may allow it within constraints. A CIO may monitor it across books. An agent can be useful only if it understands not just the number, but the context of the number. Is the exposure expected? Is it hedged? Was it approved? Does it violate a policy? Has the thesis changed? Is there a liquidity issue? Was the risk generated by a deliberate portfolio construction decision or by drift?
That requires integration between portfolio data, risk models, decision logs, approvals, and human intent.
Execution creates another challenge. An execution agent cannot simply be allowed to place trades because a model produced a recommendation. It needs permissions, pre-trade checks, trade intent, restricted lists, risk limits, liquidity constraints, approval workflows, and kill switches. It must also produce logs that can be audited later.
Reporting is similar. A reporting agent can draft a beautiful investor letter, but if it cannot reconcile performance attribution, position changes, thesis evolution, risk events, and the actual decision history, it may produce polished fiction. In institutional finance, polished fiction is worse than an ugly spreadsheet.
This is why AI-native requires process redesign. The firm must make explicit what was previously implicit. McKinsey's research on scaling agentic AI confirms this at the infrastructure level: eight in ten companies cite data limitations as the primary roadblock to scaling AI agents. The barrier is not model capability. It is that organizational data remains fragmented, siloed, and ungoverned, reducing reliability and increasing operational risk.
The Primitives Agents Need
For agents to operate effectively inside an investment firm, several technical and organizational primitives are required.
Context. Agents need access to the right information at the right time. In a hedge fund, context includes market data, fundamentals, filings, transcripts, broker research, internal notes, portfolio holdings, risk exposures, historical decisions, constraints, and client or mandate requirements. If the agent sees only one slice of the firm, it will reason from partial reality.
Memory. Investment decisions unfold over time. A fund needs to remember why it owned something, what assumptions mattered, what changed, what was debated, and how conviction evolved. Without memory, agents are condemned to operate transactionally. They can answer questions, but they cannot understand the institutional narrative.
Tool use. Agents need tools, but tool access must be bounded. There is a large difference between an agent that can read a filing, an agent that can update a research note, an agent that can propose a trade, and an agent that can execute one. Gartner's May 2026 research makes this point with striking clarity: applying uniform governance across all AI agents leads to enterprise failure. They recommend a proportional governance approach that classifies agents across distinct autonomy levels, with each level representing a different trust boundary and corresponding governance requirements. Their prediction: by 2027, 40 percent of enterprises will demote or decommission autonomous AI agents due to governance gaps discovered only after production incidents.
Identity and permissions. A human employee has a role, entitlements, manager, compliance obligations, and an audit trail. Agents need the same. Who created the agent? On whose behalf is it acting? What can it access? What can it change? What requires human approval? What happens when the agent is wrong?
Workflow state. Investment work is not just a collection of documents. It is a sequence of states: idea proposed, research in progress, thesis challenged, risk reviewed, position approved, portfolio adjusted, monitoring active, thesis impaired, exit considered. Agents need to know where they are in the workflow.
Observability. A firm must be able to see what agents are doing. What did they access? What did they infer? What actions did they recommend? Which actions were accepted or rejected? Where did they fail? This is especially important because agentic systems can initiate multi-step actions, interact with tools, and produce outcomes that are difficult to reconstruct after the fact.
Governance. Agents need policies, escalation paths, approval gates, and incident response. McKinsey's infrastructure research emphasizes that scaling agentic AI requires turning unstructured data into governed, reusable assets that systems can interpret and trust. This is not optional in financial services, where regulatory expectations around explainability and auditability are non-negotiable.
Auditability. Operational due diligence teams and regulators will not be satisfied with "the AI said so." A fund must be able to show how AI was used, what controls existed, and how decisions were made.
Human accountability. AI-native does not mean responsibility disappears into the machine. In fact, agentic systems may require more explicit accountability because the machine can act across workflows faster than humans can manually supervise.
These primitives are not naturally produced by today's fragmented hedge fund stack.
The Integrated Investment Workflow: What Agent-Ready Actually Looks Like
The primitives above are abstract. To make them concrete, consider what happens when you trace a single investment idea from inception to LP reporting. In a traditional fund, this workflow crosses seven or eight distinct systems, each with its own data model, access controls, and operational cadence. For AI agents to participate meaningfully, these systems must be connected through a shared operating layer with machine-readable state at every boundary.
Data pipelines and market infrastructure. Everything begins with data. A fund ingests market data (OHLCV bars, tick data, corporate actions), fundamental data (financial statements, valuation metrics, sector classifications), alternative data (satellite imagery, web traffic, sentiment), and reference data (security master, corporate hierarchies, index constituents). In most funds, these arrive through separate vendors with different schemas, delivery schedules, and quality guarantees. Hedge funds spent an estimated $2.8 billion on alternative data alone in 2025, a 17 percent increase from the prior year. For an AI agent to reason about a stock, it needs a unified, time-consistent view across all these sources, not a patchwork of CSVs and API calls. The data layer must provide versioned, point-in-time snapshots so that agents never inadvertently use future information, and must surface data lineage so that any downstream conclusion can be traced back to its source.
Idea generation and research. Research is where most funds first deploy AI, but the opportunity is far larger than summarization. An agent-ready research workflow captures the full lifecycle of an investment thesis: the initial signal or catalyst, the supporting evidence, the counterarguments considered, the data sources consulted, the confidence level, and the decision owner. When an agent identifies a potential opportunity (a filing anomaly, a factor dislocation, a supply chain disruption), that idea must be represented as a structured object with metadata, not buried in a notebook or chat thread. This allows other agents to monitor the thesis over time, flag contradicting evidence, and surface the idea's current status to portfolio construction without requiring a human to manually update a spreadsheet. NVIDIA's recent work on multi-agent financial signal discovery demonstrates the pattern: separate agents for signal identification, code generation, and evaluation, each operating on shared structured state with full audit trails.
Portfolio construction and optimization. Portfolio construction translates research conviction into target weights subject to constraints: risk budgets, factor exposures, sector limits, liquidity requirements, correlation targets, and drawdown thresholds. In most funds, this happens in a separate system (often a proprietary optimizer or a risk platform like Axioma or Barra) that has limited visibility into the research process that produced the conviction scores. An integrated operating layer connects research confidence directly to the optimizer's inputs, so that when a thesis is downgraded or a risk limit is tightened, the portfolio construction engine can immediately reflect the change. Agents operating in this layer can propose rebalancing trades, simulate the impact of adding or removing a position, and flag when the current portfolio has drifted from its intended construction without waiting for a human to manually run the optimization.
Execution and order management. Once target weights are determined, the execution system must translate them into orders, route those orders to the appropriate venues or brokers, manage fills, handle partial executions, and track slippage. Order management systems (OMS) are among the most mature components of the hedge fund stack, but they are rarely connected to the research and portfolio construction layers in a machine-readable way. An agent-ready execution layer knows the intent behind every order: which thesis it serves, what risk limit it satisfies, what urgency it carries, and what governance approval authorized it. This matters because execution agents need to make real-time decisions (split an order, delay for liquidity, route to a dark pool) that depend on context the OMS alone does not have. It also matters for post-trade analysis: understanding whether execution quality degraded because of market conditions, routing decisions, or timing relative to information events.
Risk management and monitoring. Risk in a hedge fund is not a single number. It is a multi-dimensional surface: factor exposures, concentration limits, liquidity risk, counterparty risk, drawdown thresholds, correlation regimes, and tail scenarios. Most risk systems operate on a snapshot basis, producing reports at end-of-day or intraday intervals. An agent-ready risk layer operates continuously, comparing the current portfolio state against declared limits and flagging breaches or near-breaches in real time. Critically, it must also understand the context of risk: an elevated factor exposure that was deliberately chosen as part of a thesis is different from one that emerged through drift. This requires integration with the research and portfolio construction layers. Risk agents can then distinguish between intentional risk-taking (which should be monitored but not blocked) and unintentional drift (which should trigger alerts or automatic rebalancing proposals).
Reconciliation and settlement. The move to T+1 settlement in the US (and the upcoming transition in Europe and the UK) has compressed the time available for trade confirmation, matching, and reconciliation from days to hours. EquiLend analysis shows that trades booked on T+2 timelines today achieve only a 65.5 percent match rate by the end of T+1, meaning settlement fails will rise sharply without automation. Reconciliation agents must compare internal records against custodian statements, prime broker reports, and counterparty confirmations, identifying and resolving breaks before they become settlement failures. This requires real-time access to trade records, position data, corporate action events, and cash movements. In a fragmented stack, reconciliation is often the most manual and error-prone process. In an integrated stack, it becomes a continuous automated function with human escalation only for genuine exceptions.
Fund administration and NAV. Fund administrators calculate net asset value, process subscriptions and redemptions, maintain the official books of record, and produce audited financial statements. Historically, the administrator operates as a separate entity with its own systems, receiving data from the fund via file transfers and manual uploads. An agent-ready administration layer maintains a continuous, reconciled view of positions, valuations, and investor balances rather than relying on periodic batch processes. This allows agents to produce real-time NAV estimates, flag valuation discrepancies before they compound, and ensure that the fund's internal records and the administrator's records remain synchronized without manual intervention.
LP reporting and investor relations. At the end of the chain, everything must be communicated to investors: performance attribution, risk exposures, portfolio changes, thesis evolution, governance events, and operational metrics. DDQ response windows have compressed from 14 days to 7, and the average questionnaire now spans 250-plus questions. An agent-ready reporting layer can assemble investor communications by pulling directly from the structured workflow state: which positions changed, why, what risk events occurred, how the portfolio construction responded, and what governance approvals were obtained. This is not about generating prettier PDFs. It is about producing reports that are internally consistent, auditable, and traceable back to the actual decision history, because that is what allocators now demand.
The integration imperative. The critical insight is that none of these components can be agent-ready in isolation. A research agent that cannot see portfolio constraints will propose ideas the fund cannot implement. A risk agent that cannot see research intent will flag every deliberate exposure as a violation. An execution agent that cannot see governance state will lack the authority to act. A reporting agent that cannot see the full decision chain will produce narratives disconnected from reality. The value of integration is not merely operational efficiency. It is the precondition for agents to reason correctly across the investment lifecycle. Each component must share a common data model, a common identity and permissions framework, a common audit trail, and a common understanding of workflow state. Without that shared substrate, agents remain trapped in silos, and the fund remains dependent on human memory to stitch the pieces together.
Fragmentation as an ODD Problem
The fragmented stack is not only a productivity problem. It is also an operational due diligence problem.
Institutional allocators do not simply underwrite returns. They underwrite organizations. They ask whether the manager can safely manage capital through stress, turnover, errors, growth, and uncertainty. They care about controls, reporting, valuation, cybersecurity, compliance, segregation of duties, business continuity, and key-person risk. In many institutional allocation processes, operational due diligence teams can veto a manager regardless of investment performance.
The numbers confirm this is not theoretical. According to Hedgeweek research, 82 percent of allocators in North America have increased their operational due diligence reviews over the past two to three years. Fifty-six percent of allocators now consider institutional-grade infrastructure a baseline expectation, and over 73 percent view the absence of an independent administrator as an immediate disqualifier. Industry surveys suggest that an estimated 85 percent of LPs rejected a manager over operational concerns alone in 2025, with DDQ response windows compressing from 14 days to 7 and the average questionnaire now spanning 250-plus questions across 21 sections.
Fragmentation makes this harder.
Suppose an allocator asks why the fund increased a position during a drawdown. In a fragmented environment, the answer may require reconstructing evidence from analyst notes, Slack messages, risk reports, portfolio history, PM recollections, broker commentary, and models. If those artifacts are scattered, the organization may still have made a good decision, but it cannot easily prove the quality of its process.
Suppose an allocator asks how AI is used in the investment process. A weak answer is that analysts use ChatGPT. A stronger answer is that AI-assisted workflows are permissioned, logged, reviewed, and auditable; that agents do not execute without approval; that outputs are linked to source data; that recommendations are reviewed by named humans; and that agent use is visible across the research, risk, and reporting lifecycle.
The second answer requires infrastructure.
This matters especially for emerging managers. Many smaller funds believe that the path to raising capital is mostly a matter of better performance. Performance matters, but institutional credibility often depends on whether the firm can demonstrate process quality. A manager with good returns but poor operational controls may fail due diligence. A manager with a credible operating system may earn conversations that would otherwise be unavailable.
This is where the opportunity becomes larger than productivity. If AI-native infrastructure can improve institutional memory, governance, workflow visibility, and auditability, it can help managers become more investable.
The False Comfort of Best-of-Breed Tools
One reason the current stack persists is that each component is rational. Bloomberg is rational. Excel is rational. Slack is rational. Python is rational. Enfusion is rational. Snowflake is rational. Fund administrators, OMS vendors, PMS vendors, risk platforms, custodians, and data providers all solve real problems.
The issue is not that these tools are bad. The issue is that the operating model they create is not agent-native.
Best-of-breed worked reasonably well when humans were the integration layer. Humans could move information between systems, interpret context, make judgment calls, and manually enforce workflows. The cost of that approach was headcount, coordination, and operational drag.
AI changes the integration problem. If agents are expected to perform multi-step work, then the firm needs a machine-readable operating layer across systems. Otherwise each agent becomes trapped inside a silo. The research agent cannot see portfolio intent. The risk agent cannot see research conviction. The reporting agent cannot see decision lineage. The execution agent cannot see governance state. The compliance workflow cannot see what the research process actually did.
This is the pattern McKinsey identifies across industries: organizations have deployed AI widely (nearly nine in ten companies use it in at least one business function) but 94 percent report not seeing significant value from those investments. The gap is not model quality. It is organizational readiness. The firms that treat AI as a tool to insert into existing workflows get incremental productivity. The firms that redesign workflows around AI-native primitives get compounding leverage.
This is why the next generation of investment infrastructure is unlikely to be just another application. It needs to be a coordination layer.
The Role of an AI-Native Operating Layer
An AI-native operating layer for investment firms would not replace every system. It would not replace prime brokers, fund administrators, custodians, market data vendors, OMS platforms, or risk models. Those systems will continue to matter.
Instead, the operating layer would sit across them. It would provide context, memory, workflow state, permissions, agent orchestration, governance, and auditability.
In practical terms, such a layer would allow a fund to connect research workflows with portfolio data, risk constraints, monitoring, execution intent, reporting, and institutional memory. It would allow agents to operate within clearly defined boundaries. It would allow humans to supervise not just outputs, but workflows. It would produce an operating record that can be used internally by PMs and externally in allocator conversations.
This is the subtle argument for Podium.
The investment industry does not need another AI chatbot. It does not need another isolated research assistant. It does not need an agent that generates trade ideas without institutional context. What it needs is an AI-native operating environment that lets lean teams behave with the discipline, memory, and coordination of much larger organizations.
The ambition is not full autonomy. The ambition is operational leverage.
A two-person team should not pretend to be Millennium. But with the right infrastructure, it may be able to operate with a degree of process discipline, monitoring, documentation, and AI-assisted scale that would have required far more people in the past. A $100 million emerging manager may not need to build an internal engineering organization just to coordinate research, reporting, and monitoring. A $500 million fund may not need to let its AI adoption devolve into dozens of disconnected pilots. A family office may not need six people around every new mandate if parts of onboarding, monitoring, reporting, and knowledge capture can be re-engineered around AI-native workflows.
That is the broader shift.
Why This Matters Now
Three forces are converging.
First, AI adoption in investment management is accelerating and allocator expectations are shifting with it. The AIMA data shows near-universal adoption among hedge funds, and the allocator signal is clear: 60 percent of institutional investors say they would be more likely to invest in a fund that commits meaningful budget to AI. This is no longer a differentiator. It is becoming table stakes.
Second, enterprise AI is moving from copilots to agents, and the governance challenge is intensifying. Gartner predicts that by 2027, 40 percent of enterprises will be forced to roll back autonomous AI agents due to governance failures. McKinsey frames the shift as a fundamental challenge to organizational operating assumptions: agentic AI is not just a tool upgrade but a participant in execution that can initiate actions, sequence work, and coordinate across functions. The firms that deploy agents without redesigning their operating model will face the same fate as the 80 percent of companies currently seeing no bottom-line impact from AI investments.
Third, the economics of investment organizations are under structural pressure. BCG's 2026 research shows that more than 80 percent of revenue growth in 2025 came from markets, not managers. Fees are falling, costs are rising, and margins have not moved in 15 years. The traditional assumption that scale automatically improves profitability has broken down. Launching and scaling a fund increasingly requires infrastructure that many talented managers cannot afford to build from scratch.
These forces point in one direction: the operating stack of investment management must change.
The firms that merely add AI tools to fragmented workflows will get incremental productivity. The firms that redesign their workflows around AI-native primitives may get something more valuable: organizational compression. They may allow smaller teams to do the work of larger teams, not because AI replaces judgment, but because AI reduces the cost of coordination, memory, monitoring, and operational execution.
This is the real opportunity. Not autonomous hedge funds. Not AI stock pickers. Not magic alpha machines.
The opportunity is to rebuild the logistics of investing.
For decades, the best investment firms won not only because they had better tactics, but because they had better machinery. In the next decade, the machinery itself will be rebuilt around agents, humans, governance, and integrated operating systems. The firms that understand this early will not ask whether AI can write a better memo. They will ask what an investment organization looks like when every workflow is designed for humans and agents from the start.
That is the future Podium is building toward.