Shadow-trading approval rail that certifies third-party AI agents before hedge funds let them touch live capital.

1. A true third-party supply layer is emerging, so funds now need software to evaluate outside agents at scale rather than reviewing one-off internal models.
2. Because buyers are being offered agents across research, risk, and execution together, the first winning control product can sit at the workflow boundary instead of inside any single point tool.
3. The launch itself bakes in human review and final sign-off, proving that trust and supervision are first-order product requirements, not afterthoughts.
4. Waton's disclosed revenue, cash position, and full strategic pivot suggest AI-native finance is getting real budget and distribution effort, which makes an enabling control layer more commercially plausible now.

Catalyst. The emergence of a ranked marketplace for third-party trading agents, combined with mandatory human review in the launch product itself, means the adoption bottleneck has shifted from finding agents to supervising them safely.

Build a shadow-agent approval rail that plugs into agent outputs, market data, and a fund's existing OMS or paper-trading stack. Before an external agent can trade live, the system runs it in shadow mode, records every idea, order, and override, and tags each action against mandate rules, concentration limits, approved instruments, and reviewer comments. The product packages those runs into a standardized approval memo with attribution, exception summaries, and side-by-side comparisons against the desk's current process. Once an agent is approved, the same rail becomes the monitoring layer for ongoing human sign-off, drift alerts, and kill switches when behavior deviates from the approved envelope.

What's different. This is not another AI stock picker and not a generic compliance wrapper. The product is opinionated around one painful moment: deciding whether an outside trading agent deserves live capital. By owning shadow runs, mandate mapping, reviewer evidence, and post-approval drift monitoring in one rail, it becomes the neutral approval system between agent marketplaces and professional investors.

Jobs to be done

flowchart LR
  Buyer[Hedge fund COO or Head of Risk] --> Pain[Cannot trust external agents with live capital]
  Pain --> Product[Shadow-trade approval rail]
  Product --> Outcome[Faster governed rollout of AI trading agents]

Executive takeaways

• Third-party agent supply is beginning to outrun allocator trust infrastructure: Waton opened a third-party Agent Talents Market but still requires mandatory human review and final sign-off [1].
• The initial buyer pool is narrow but real: hedge fund capital is at a record $5.22T, AIMA serves 2,000+ alternative-manager members, and Campden regional family-office averages sit squarely in the $500M-$5B AUM band [2][4][6][7][8].
• Adoption risk is operational and governance-heavy, not conceptual: 72% of surveyed APAC buy-side firms already use AI moderately, while AIMA and regulators emphasize shadow AI, vendor risk, testing, records, and human oversight [11][12][13][14][15][17][19].
• The market is crowded with adjacent workflow and surveillance products, but no clear incumbent owns neutral pre-live certification of third-party trading agents; the biggest threats are in-house builds and platform bundling [27][28][29][30][36][37][38][39].

Market definition

This category is a pre-live certification and live-supervision control layer for external AI trading agents used by hedge funds and institutional family offices. It sits between agent marketplaces or brokers and the fund’s OMS or execution stack, converting shadow runs, rule checks, overrides, and post-approval drift into a committee-ready approval record [1][14][15][17][19][31][33][34].

Customer and buyer

The economic buyer is usually the COO, Head of Risk, or COO-adjacent operating principal who owns vendor diligence, paper-trading evidence, and rollout approval. Daily champions sit in quant platform, desk operations, or trading-technology teams, while PMs and quant leads are the internal demand creators who discover promising agents but cannot approve live capital alone [12][13][15][16][21][22].

Buying triggers

• A PM or quant lead finds a promising external agent or broker algo, but risk requires a repeatable shadow record before live capital is approved. [1][24][25]
• Volatility and market-impact pressure push funds toward automation, increasing demand for better algo evaluation and post-trade evidence. [23][25][26]
• Vendor outsourcing and algorithmic-trading controls make ad hoc spreadsheets and screenshots hard to defend in diligence or exams. [14][15][16][17][19][21]

Willingness to pay

Adjacency budget already exists inside risk-tech, OMS or OEMS, algo-monitoring, and institutional quant infrastructure. Public materials from QuantConnect, FlexTrade, Eventus, and Limina show buyers already fund research or execution, workflow, and surveillance stacks, so the startup can sell as an added control plane rather than a net-new science project. [29][30][37][39]

Category dynamics

Growth signal 4.2% HNWI wealth growth in 2024

Validation signals

• Waton launched an Agent Talents Market with mandatory human review and final sign-off.
• 72% of surveyed APAC buy-side firms already use AI moderately and 66% prioritize workflow automation.
• Hedge funds cite reduced market impact as the top reason to use algos during volatile markets.
• QuantConnect publicly halted Alpha Streams 1.0 because poor out-of-sample performance made the marketplace hard to justify.
• Institutions already buy post-trade transparency and regulatory analytics around algos, which supports willingness to pay for a better pre-live layer.

Regulatory & technical constraints

• The rail must preserve replayable order and recommendation logs plus enough records for competent-authority review.
• Funds and their brokers will expect pre-set trading and credit thresholds plus restricted access, not just post-hoc dashboards.
• External agents trigger vendor due-diligence, contracting, and ongoing-monitoring obligations even when the startup is only the control layer.
• Paper trading alone omits market impact, information leakage, slippage, queue position, and some fees, so certification needs conservative stress assumptions.
• Post-approval monitoring should cover both runtime behavior and data or output drift if the startup wants to support live supervision credibly.

Direct competition is still thin at the exact pre-live certification layer. Waton, QuantConnect, and Darwinex focus on agent or strategy supply and capital-allocation workflows, while Eventus, TradingHub, FlexTrade, and Limina cover surveillance or order workflow. The white space is a neutral approval rail that scores third-party agents before go-live and keeps a live drift envelope afterward [1][28][29][30][36][37][38][39].

Why incumbents do not win by default

• Agent marketplaces and workbenches. Waton, QuantConnect, and Darwinex optimize for agent supply, track-record creation, or on-platform workflow, but allocators still need neutral approval evidence when creator logic stays opaque or incentives favor marketplace growth over buyer diligence.
• OMS and OEMS vendors. FlexTrade and Limina own order and exception workflow plumbing, yet they do not decide whether a third-party agent deserves capital before live deployment.
• Surveillance vendors. Eventus and TradingHub monitor behavior after or around deployment, but they are not designed to produce investment-committee approval packets for new external agents.
• Cloud and agent tooling. LangChain, cloud monitoring tools, and audit-log products provide HITL, tracing, and control primitives, but not fund-specific mandate mapping, broker-rule alignment, or allocator diligence workflows.

Shadow Agent Approval Rail sells a pre-live certification and live-supervision control plane to U.S. hedge funds and institutional family offices evaluating third-party AI equities agents. The first product runs one external agent in read-only shadow mode, maps each recommendation, order, and override to the fund's mandate and risk rules, and produces an investment-committee-ready approval packet for a single PM pod. The first customer is a $500M-$5B AUM multi-manager hedge fund or institutional family office where a PM has found a promising outside agent but the COO or Head of Risk will not approve live deployment without repeatable evidence. This wedge is attractive because the same operating buyer already owns vendor diligence, paper-trading evidence, and ongoing monitoring, letting the company land with certification and expand into live drift monitoring instead of trying to replace the OMS or become a strategy marketplace. Research supports a narrow but real launch market of about $51M SAM and a modeled first-year ACV centered around $200k, but both figures depend on proving that enough target funds are actively evaluating external agents now. The company should start with U.S. long-short equities, read-only integrations, and one-pod pilots, then add deeper OMS hooks and broader monitoring only after approval-to-production conversion is repeatable. The moat is accumulated approval history linking recommendations, fills, overrides, mandate exceptions, committee outcomes, and post-approval drift envelopes across many agent evaluations. The key disconfirming risk is category timing: if most target funds still prefer internal agents or are not yet testing external supply, the business stays niche and should not scale hiring ahead of proof.

Problem

• Funds can now discover third-party AI trading agents, but risk and operations teams still lack a defensible way to decide whether an external agent deserves live capital.
• Current approval work is manual paper trading, OMS notes, spreadsheets, and screenshots, which are slow to review, hard to compare across agents, and weak under vendor or regulatory scrutiny.
• Paper trading alone misses slippage, market impact, queue position, and drift risk, so buyers can approve the wrong agent or block deployment indefinitely.

Solution

• Ingest agent recommendations, fills, and human overrides in read-only shadow mode for one U.S. equities pod, then map every proposed action to mandate rules, position limits, approved instruments, and reviewer comments.
• Generate a committee-ready approval packet with replay, attribution, exceptions, conservative paper-to-live stress assumptions, and an explicit approve or reject decision trail.
• After approval, keep the same rail in place for live drift monitoring, threshold alerts, and kill-switch workflows so certification and supervision stay in one system.

Why we win

• The product sits at the exact approval decision before live capital, where neither agent marketplaces nor surveillance vendors are neutral or optimized today.
• By starting read-only and above funds flow, the company can deliver value before deep OMS or broker changes, reducing integration and regulatory friction in the first sale.
• Each certification compounds a proprietary dataset of shadow behavior, mandate exceptions, reviewer decisions, and approved drift envelopes that becomes harder for in-house scripts to match across many agents.

Milestones

flowchart LR
  Wedge[One-agent approval wedge] --> MVP[Read-only certification MVP]
  MVP --> Proof[Approved agents and referenceable packets]
  Proof --> Expansion[Live monitoring and multi-pod expansion]

Founding team

Experiment roadmap

Risk assessment

What must be true

• At least 5 of the first 15 target funds are actively evaluating third-party agents today rather than only internal models.
• The MVP can produce an approval packet from read-only exports without requiring order-routing changes in the first deployment.
• Target buyers accept at least $150k of first-year ACV when the workflow replaces weeks of manual due diligence and supports one-pod go-live.
• At least 50% of paid certifications convert into live-monitoring subscriptions because the same buyer owns post-approval supervision.
• Neutral cross-platform certification wins head-to-head against marketplace-native or in-house workflows in at least 3 early competitive evaluations.

Open diligence questions

• How many U.S. hedge funds in the target AUM band are already evaluating external agents versus only internal models?
• What minimum dataset does an investment committee require to approve an external agent without deep OMS integration?
• Who actually controls budget and vendor approval in practice: COO, Head of Risk, CTO, or desk-platform lead?
• How often do marketplaces, brokers, or OMS vendors prefer to partner rather than bundle their own approval workflow?
• What evidence convinces buyers that paper-to-live slippage and drift are handled conservatively enough to trust the certification result?

Model sanity

• Revenue engine. Base-case Y3 revenue comes from growing from 4 to 18 paying funds while lifting blended ARR from pilot-heavy Y1 levels to about $250K once monitoring attaches.
• Must go right. The company must keep certification-to-monitoring conversion near the plan's 50%+ target and make read-only integrations repeatable enough that deployment time keeps shrinking.
• Model breaks if. A one-quarter slip in buyer timing or a monitoring attach rate closer to the downside case removes more than $0.5M of Y3 revenue and pushes cash meaningfully lower.
• Next-round proof. Reaching roughly 10 paying funds by Q4Y2 with at least one credible partner channel is the operating proof that justifies the next financing.

Scenarios

Sensitivity

flowchart LR
  TargetAccounts --> PaidPilots
  PaidPilots --> CertifiedFunds
  CertifiedFunds --> MonitoringAttach
  CertifiedFunds --> Revenue
  MonitoringAttach --> Revenue
  Revenue --> GrossProfit
  GrossProfit --> Cash

Flags: The launch market is intentionally narrow, so one or two delayed hedge-fund decisions can move the model materially even though Y3 logo count is still below 10% of the researched SAM. · Gross margin only clears the 70% target after reusable integrations reduce implementation drag; if read-only data proves insufficient and deeper OMS work is needed, this model is too optimistic on burn. · The revenue plan assumes monitoring expansion lifts blended ARR from roughly $234K in Y2 to about $250K in Y3; if customers buy certification only, the base case overstates revenue density. · No debt, capex, or unusual regulatory-compliance project spend is modeled, so the funding ask would need to rise if a customer requires bespoke control development outside the roadmap.

• Category timing risk. If third-party trading-agent marketplaces grow slower than expected, the initial buyer pool may stay niche for too long. Mitigation: Start with funds already paper-trading external agents and support internal agent approval flows so the product does not depend on one marketplace alone.
• Integration friction. Funds may resist a new system if it requires deep execution-stack changes before value is visible. Mitigation: Begin in read-only shadow mode using agent outputs and paper-trading feeds, then add deeper OMS and live-monitoring hooks after approval workflows prove ROI.
• In-house build temptation. Sophisticated funds may believe internal quants can stitch together paper trading, rule checks, and review logs themselves. Mitigation: Win with faster time to approval, cleaner reviewer evidence, and a reusable cross-agent control layer that is hard to maintain as agent volume grows.