AIG-McGill $1.6B Agentic AI Deal Reshapes the Follow Market

On March 16, 2026, AIG and McGill and Partners announced a collaboration that commits up to $1.6 billion in gross premiums written of AIG capacity across McGill’s specialty portfolio. AIG will deploy 25% of its capacity through McGill’s digital broking platform, with agentic AI capabilities managing the allocation of that capacity to individual risks. The technology backbone is Palantir’s Foundry platform, which constructs an ontology of McGill’s entire portfolio, providing near real-time exposure data, limit deployment, modeled risk outputs, and loss information.

From tracking AIG’s patent filings and earnings disclosures across eight quarters, the shift from internal LLM tools to externally deployed agentic capacity marks a distinct strategic inflection. AIG is no longer just using AI to process submissions faster internally. It is deploying autonomous AI agents to make follow underwriting decisions on live specialty risks, in partnership with a broker that has already built digital capacity agreements with AEGIS London and AXA XL.

Most coverage has treated this as a standard broker-carrier partnership announcement. It is not. This deal introduces algorithmic follow capacity at a scale that could structurally change how the London subscription market functions, with direct implications for actuarial pricing, portfolio management, and the employment pipeline for junior specialty underwriters.

What the Deal Actually Looks Like

The structure of the AIG-McGill collaboration has several components worth unpacking separately, because the headline number alone understates how the deal actually works.

Capacity commitment: AIG commits capacity equal to 25% across up to $1.6 billion of McGill’s specialty gross premiums written. This is not a single program; it spans McGill’s full specialty portfolio. Peter Zaffino, AIG’s Chairman and CEO, described it as a “significant opportunity to deliver greater efficiency to the subscription market.”

Underwriting criteria: AIG developed underwriting criteria that enable real-time underwriting through McGill’s digital broking platform. The criteria are embedded in the platform, meaning the AI system can evaluate whether a risk fits AIG’s appetite and allocate capacity without routing every decision through a human underwriter.

Palantir Foundry ontology: AIG collaborated with Palantir to build what they call an “ontology” of McGill’s entire portfolio using Palantir’s Foundry platform. In practical terms, this means every entity, risk, exposure, and relationship in the portfolio is mapped into a structured digital representation that LLM agents can query. The system provides near real-time data on exposures, limit deployment, modeled risk outputs, and loss information.

Agentic AI deployment: McGill will leverage agentic AI capabilities to manage the deployment of AIG’s capacity to clients. This is the critical distinction from traditional automated underwriting. The AI agents do not just score risks and hand recommendations to humans. They orchestrate the entire follow underwriting workflow: ingesting broker submissions, evaluating them against AIG’s criteria, checking portfolio-level exposures through the Foundry ontology, and allocating capacity.

Steve McGill, the firm’s CEO and founder (formerly Group President of Aon plc), put it directly: “This collaboration has the potential to disrupt the dynamics of the subscription market.”

How the Subscription Market Works, and Where AI Fits

To understand why this deal matters, you need to understand how the Lloyd’s subscription market actually functions.

A broker prepares a “slip,” formally known as a Market Reform Contract, that summarizes the risk, proposed terms, premium, and total limit. The broker then approaches a lead underwriter, who negotiates the final terms, sets the premium rate, and signs the slip for a portion of the risk, typically 15% to 20%. The lead underwriter does the deep analysis: reviewing the submission data, challenging the broker on exposure details, negotiating exclusions, and ultimately setting the price that every other participant on the slip will follow.

Follow underwriters then subscribe to the remaining capacity at the same terms the lead set. The follow decision is ostensibly independent, but in practice, follow underwriters rely heavily on the lead’s assessment. The economics of follow underwriting are straightforward: you accept the lead’s pricing in exchange for lower acquisition costs and reduced analysis burden. Claims are managed by the lead syndicate on behalf of all participants.

This system has worked for over three centuries. It is also slow, relationship-driven, and labor-intensive. A single specialty placement can take days or weeks as brokers physically or electronically circulate the slip among potential followers. The information asymmetry between lead and follow is real but tolerated because the follow market provides essential capacity diversification.

Algorithmic follow underwriting changes this dynamic fundamentally. If an AI system can evaluate a risk against predefined criteria, check portfolio-level correlations in real time, and commit capacity within minutes, the traditional follow underwriter’s role becomes harder to justify economically. The AI follow system does what human follow underwriters do: it trusts the lead’s pricing judgment and evaluates whether the risk fits appetite. It just does it faster, cheaper, and with continuous portfolio-level awareness that no individual human underwriter can maintain.

Ki Insurance Set the Precedent; AIG-McGill Scales It

The concept of algorithmic follow underwriting is not new. Ki Insurance, launched in May 2020 as a partnership between Brit Insurance and Google Cloud, was the first fully algorithmic follow-only Lloyd’s syndicate. Ki does not write lead business. It evaluates risks that have already been led by other syndicates and decides algorithmically whether to follow.

Ki’s results through 2025 validate the model. The syndicate posted record results: $1.11 billion in gross managed premium (up 6.9% year over year) and $171.4 million in profit before tax in its first year as a standalone entity after separating from Brit effective January 1, 2025. Capacity partners include Beazley, QBE, Aspen (Sompo), and Travelers. QBE has deepened its commitment by investing in Aurora, an algorithmic MGA that extends the same concept into delegated authority. TMK joined as a capacity partner in April 2026.

Ki proved that algorithms can follow underwrite profitably. The AIG-McGill deal scales the concept in two important ways.

First, the capacity commitment is larger. Ki manages approximately $1.1 billion in gross managed premium. The AIG-McGill deal covers up to $1.6 billion in gross premiums written, and that represents only 25% of AIG’s capacity flowing through this single broker relationship. The total AIG capacity accessible through this arrangement is substantially larger than Ki’s entire operation.

Second, the technology architecture is more sophisticated. Ki uses Google Cloud’s infrastructure for risk evaluation. The AIG-McGill system uses Palantir’s Foundry, which constructs a structured ontology of the entire portfolio, mapping entities, exposures, relationships, and loss data into a queryable digital representation. This enables the agentic AI layer to make allocation decisions informed by portfolio-level context, not just individual risk evaluation. When AIG’s system evaluates a new submission, it can simultaneously check how that risk correlates with the existing 75% of AIG’s portfolio that is managed through other channels.

What “Agentic AI” Actually Means in Underwriting

The term “agentic AI” has become a marketing fixture in insurance technology. To separate the substance from the branding, it helps to draw a clear boundary between what traditional ML scoring does and what agentic systems do.

Traditional ML scoring models take structured inputs and produce a single output: a risk score, a price indication, a classification. The human underwriter receives the score and decides what to do with it. The model has no awareness of the broader workflow, no ability to request additional information, and no capacity to coordinate with other systems.

Agentic AI systems are autonomous multi-agent architectures that reason, decide, and execute across complex workflows with minimal human direction. In underwriting, this means:

• Intake agents extract data from unstructured broker submissions (PDFs, spreadsheets, emails), verify it against internal and third-party data sources, identify missing fields, and proactively communicate with brokers to fill gaps. Industry benchmarks show 92% to 94% extraction accuracy for insurance-specific entities using NLP combined with computer vision.
• Evaluation agents apply underwriting criteria, check exposure concentrations against portfolio-level data, run modeled risk outputs, and produce a recommendation. Some architectures employ adversarial self-critique, where a “critic agent” challenges the primary agent’s conclusions before the recommendation is submitted.
• Orchestration agents coordinate the entire workflow end to end, routing risks through the appropriate evaluation pipelines, managing exceptions that require human review, and updating portfolio analytics in real time as capacity is deployed.

Performance benchmarks from early adopters are striking. Hiscox has reported a 99.4% reduction in quote cycle time for London Market specialty risks, from three days to approximately three minutes. Across the industry, straight-through processing rates have increased from 10% to 15% in traditional workflows to 70% to 90% with agentic systems. Loss ratio improvements of three to five percentage points have been reported by carriers using advanced underwriting AI, which on a $1 billion premium portfolio translates to roughly $30 million to $50 million in annual profit improvement.

The actuarial judgment boundary sits at the point where the agentic system encounters ambiguity that its training data and ontology cannot resolve. For standard follow risks with well-understood exposure profiles, that boundary is far from the initial submission. For novel risks, complex casualty placements with unusual coverage terms, or heavily accumulated catastrophe exposures, the boundary is closer. The AIG-McGill system appears designed to handle the former category autonomously while routing the latter to human underwriters and actuaries.

AIG’s Broader GenAI Rollout Provides the Foundation

The McGill collaboration did not emerge in isolation. AIG has been building toward this moment through a systematic GenAI rollout that accelerated through 2025 and into 2026.

AIG Assist (formally “Underwriting by AIG Assist”): AIG’s GenAI tool ingests, reviews, and prioritizes submissions before underwriters arrive at their desks. By the Q4 2025 earnings call in February 2026, AIG had expanded it to seven additional lines of business including Lexington, with accelerated rollout across remaining North America, UK, and EMEA commercial lines moved up by six months.

Lexington submission volume: The results at Lexington, AIG’s excess and surplus lines carrier, quantify the impact. Submission count increased 26% year over year. The middle market property submit-to-bind ratio increased 35%. AIG processed over 370,000 submissions by end of 2025, already at 74% of its original 500,000-submission target set for 2030.

Zaffino on capacity without headcount: CEO Peter Zaffino described the outcome bluntly on the Q4 2025 call: “We’re seeing a massive shift in our ability to process a significant submission flow way beyond our expectations without additional human capital resources. That has been the biggest surprise.” He added: “The acceleration and the opportunity is greater than I thought at Investor Day.”

Technology investment: AIG has spent approximately $300 million on data, digital workflow, AI, and talent over the past two years, with over $1 billion invested in foundational data technologies over the past five years. The Palantir partnership is central: AIG uses Foundry to build what it calls a “comprehensive digital twin” of its processes, workflows, and data.

Syndicate 2479 as a proving ground: In December 2025, AIG partnered with Amwins and Blackstone to launch a $300 million premium Lloyd’s syndicate managed by Talbot Underwriting. Palantir’s Foundry was used to construct the portfolio ontology, accessing over four million industry data points. Multiple LLM agents evaluate whether program portfolios align with the syndicate’s risk appetite by examining individual risk characteristics. This syndicate effectively served as a controlled test of the same technology now deployed at larger scale through the McGill collaboration.

Claude Wade, AIG’s EVP overseeing the AI transformation, described the vision as enabling one underwriter to do the work of five: “Review 100 percent of every private and non-profit business submission that comes in, without adding underwriters.” Data collection accuracy improved from approximately 75% to upward of 90%.

How Peers Are Responding

AIG is not the only carrier moving aggressively on AI. But the approaches differ in ways that reveal distinct strategic bets about where value will accrue.

Travelers and Anthropic: In January 2026, Travelers announced a partnership with Anthropic deploying personalized Claude and Claude Code AI assistants to nearly 10,000 engineers, data scientists, analysts, and product owners. Beyond this core team, over 20,000 Travelers employees use AI tools regularly through TravAI, the carrier’s in-house agentic AI platform. CTO Mojgan Lefebvre described the philosophy as “buy for commodity, build for advantage.” Kate Jensen, Anthropic’s Head of Americas, called it “exactly where applied AI is headed: personalized, context-aware and integrated with the systems people already use.”

The Travelers model is fundamentally different from AIG’s. Travelers uses AI to augment human productivity across existing workflows. AIG is using AI to replace human decision-making in follow underwriting for eligible risks. Both approaches work, but they imply very different workforce trajectories.

Chubb’s automation-driven headcount reduction: In its December 2025 investor presentation, Chubb disclosed plans for a workforce reduction of up to 20% over three to four years, approximately 8,600 of 43,000 global employees. The targets: 85% of major underwriting and claims processes automated, 85% of global GWP generated by fully digital or significantly digitally enabled businesses. Expected run-rate expense savings are equivalent to roughly 1.5 combined ratio points. Chubb’s approach is to reduce headcount primarily through attrition while investing in engineering, with 3,500+ engineers globally and hubs in Mexico, Greece, India, and Colombia. CEO Evan Greenberg framed AI’s potential as “breathtaking” but cautioned that “technology is evolving but human nature has not evolved.”

These three approaches form a spectrum:

Carrier	AI Strategy	Workforce Impact	Key Technology Partner
AIG	Agentic follow underwriting deployed externally	Process more volume without adding staff	Palantir (Foundry), Anthropic (Claude)
Travelers	Foundation model assistants for existing staff	Augment 20,000+ employees with AI tools	Anthropic (Claude, Claude Code)
Chubb	End-to-end process automation	20% headcount reduction over 3-4 years	In-house engineering (3,500+ engineers)

AIG’s approach is the most externally visible because it deploys agentic AI into the subscription market itself, where other participants can observe its effects. When AIG’s algorithmic follow capacity accepts or declines a risk through McGill’s platform, that decision is visible to brokers and other syndicates. Travelers’ AI deployment is largely internal. Chubb’s is structural but gradual.

Actuarial Implications of Algorithmic Follow Capacity

For pricing actuaries, the rise of algorithmic follow underwriting changes several foundational assumptions.

Speed-to-bind compresses adverse selection windows. In traditional subscription markets, there is an information gradient between the lead (who has spent days analyzing the risk) and follows (who may see the slip days later). Adverse selection operates in that gap: the worst risks get filled fastest because brokers push hardest to close placements before new information surfaces. When algorithmic follows can evaluate and commit within minutes, the information gap narrows. This should, in theory, improve the risk quality of followed business, but it also means that risks that algorithms consistently decline may concentrate in the residual human-follow pool.

Portfolio correlation monitoring becomes continuous. The Palantir ontology provides AIG with real-time visibility into how each new risk correlates with the existing portfolio. This is a fundamental change from traditional follow underwriting, where portfolio-level exposure management happens quarterly or at renewal. Continuous portfolio monitoring should reduce accumulation risk, but it also means that the pricing of individual risks is increasingly influenced by portfolio-level factors that the lead underwriter may not see.

Loss development patterns may change. If algorithmic follow underwriting improves risk selection, actuaries should expect to see different loss development patterns on algorithmically followed business versus human-followed business over the next several accident years. This will create a segmentation challenge in reserving: how do you develop separate loss triangles for algorithmically selected versus traditionally selected risks when the underlying coverage and pricing terms are identical?

Expense ratio implications are significant. Follow underwriting has always been cheaper to operate than lead underwriting because the analytical burden is lower. Algorithmic following makes it cheaper still. If AIG can process $1.6 billion in follow capacity through McGill with minimal human underwriter involvement, the expense ratio on that book should be materially better than traditionally staffed follow business. This gives algorithmic followers a structural cost advantage that can be deployed as either margin or competitive pricing.

Labor Market Effects: What Happens to Follow Underwriters

The subscription market employs thousands of follow underwriters across Lloyd’s syndicates, company market offices, and specialty reinsurers. These roles typically involve reviewing slips that leads have already priced, checking that risks fit the syndicate’s appetite, and signing lines. The analytical intensity is lower than lead underwriting, and the decision framework is more constrained.

That profile makes follow underwriting particularly susceptible to automation. The AIG-McGill deal is not eliminating follow underwriters today, but it demonstrates that the work they do can be performed by agentic AI systems operating through digital broking platforms. If the model works at $1.6 billion, there is no technical barrier to extending it.

The labor market effects will likely unfold in three phases:

Phase 1 (2026-2027): Parallel operation. Algorithmic follow capacity operates alongside human follow capacity. Brokers route straightforward placements through digital platforms while continuing to use traditional placement for complex or unusual risks. Human follow underwriters still have jobs, but submission volumes through their desks plateau.

Phase 2 (2028-2029): Volume shift. As algorithmic follow systems demonstrate consistent profitability and brokers become comfortable with digital placement, a larger share of standard specialty business flows through automated channels. Junior follow underwriter hiring slows. Syndicates that have not invested in algorithmic capability find it harder to attract capacity partners.

Phase 3 (2030+): Structural change. Follow underwriting for standard specialty risks becomes predominantly algorithmic. Human follow underwriters specialize in complex, novel, or heavily negotiated placements where the lead’s terms require adaptation. The total number of follow underwriting positions declines, but the remaining roles carry higher analytical demands and compensation.

For actuaries, the implications are nuanced. Reserving and pricing actuaries who support follow underwriting teams will see their roles evolve from traditional reserve analysis toward model validation and algorithm governance. The actuarial skillset shifts from “review loss development factors on the followed book” toward “validate that the algorithmic follow system’s risk selection criteria produce acceptable loss ratios within the expected confidence interval.”

Regulatory and Governance Considerations

Algorithmic follow underwriting raises regulatory questions that the current framework is not fully equipped to answer.

The EU AI Act, phasing in through 2026, classifies insurance underwriting AI as “high-risk,” requiring documentation, human oversight, bias testing, and explainability. An agentic system that autonomously commits capacity to individual risks will need to demonstrate compliance with these requirements. Several US states are considering mandatory algorithmic audits for underwriting AI. The NAIC’s Spring 2026 panel on agentic AI identified specific governance gaps: current model risk management frameworks were designed for traditional ML scoring models, not for autonomous multi-agent systems that make binding decisions.

For the AIG-McGill system, several governance questions are particularly relevant:

• Accountability for follow decisions: If the agentic AI commits AIG capacity to a risk that produces a large loss, who bears responsibility for the underwriting decision? The traditional follow market assigns responsibility to the follow underwriter who signed the line. Algorithmic following complicates this.
• Model validation scope: ASOP No. 56 requires actuaries to assess models used in actuarial work. When an agentic AI system autonomously commits capacity based on its evaluation of a Palantir ontology, the scope of “model validation” expands dramatically. The actuary is no longer validating a single predictive model; they are validating an orchestrated system of agents, data feeds, and decision rules.
• Transparency to co-subscribers: Other syndicates that follow on the same slip may not know that AIG’s follow decision was made by an algorithm rather than a human underwriter. Should there be disclosure requirements for algorithmic participation in the subscription market?

These are not hypothetical concerns. As algorithmic follow capacity grows from Ki’s $1.1 billion and AIG-McGill’s $1.6 billion toward potentially tens of billions in automated follow capacity across the London market, regulators will need to address them. Lloyd’s itself has been a facilitator of digital placement through its electronic placement support, but the regulatory framework has not caught up to the reality of agentic AI making binding underwriting decisions.

The Palantir Ontology as a Competitive Moat

A detail worth isolating is the role of Palantir’s Foundry in creating a structural advantage that competing carriers cannot easily replicate.

Palantir’s Ontology is an operational layer that sits atop an organization’s integrated digital assets and maps them to real-world counterparts. In the AIG-McGill context, this means every policy, risk, counterparty, exposure, and loss in McGill’s portfolio is represented as an object in a structured, queryable graph. The ontology defines object types and their relationships, then LLM agents can query this representation to perform complex evaluations that would previously require manual underwriter analysis.

AIG has invested over $1 billion in foundational data technologies over five years to enable this kind of integration. The Syndicate 2479 deployment in December 2025, which accesses over four million industry data points through Foundry, demonstrated the architecture at a smaller scale. Palantir CEO Alex Karp noted that “AIG’s deployment helps drive new partnership opportunities and efficiencies.”

For carriers that want to replicate what AIG is doing, the barrier is not the AI models themselves. Foundation models from Anthropic, OpenAI, and Google are commercially available. The barrier is the data infrastructure: the years of work required to integrate disparate systems into a unified representation that AI agents can query in real time. Carriers that have not started this data integration work are several years behind, and the gap widens as AIG and Palantir accumulate operational data from live algorithmic underwriting decisions.

What to Watch Through 2026

Loss ratio performance of the McGill book. The ultimate test of algorithmic follow underwriting is whether it produces acceptable loss ratios. AIG will not disclose the McGill collaboration’s results separately, but analysts should watch for commentary on Q2 and Q3 2026 earnings calls about specialty segment performance and digital underwriting metrics.

Broker adoption rates. McGill has already built digital capacity agreements with AEGIS London and AXA XL. If the AIG collaboration succeeds, expect other major brokers (Marsh, Aon, WTW) to accelerate their own digital placement platforms. The brokers that control digital placement infrastructure will capture disproportionate value as the follow market automates.

Ki’s continued expansion. Ki added TMK as a capacity partner in April 2026 and continues to grow. If both Ki and the AIG-McGill system perform well, the case for algorithmic follow underwriting becomes increasingly difficult for other syndicates to ignore.

Regulatory response at Lloyd’s. Lloyd’s has been supportive of digital placement, but the governance framework for algorithmic underwriting decisions has not been formally addressed. Watch for guidance from Lloyd’s on disclosure requirements, model governance standards, and accountability frameworks for algorithmic follow capacity.

Junior specialty hiring data. The pipeline effect will be subtle initially. Watch for changes in graduate scheme placements, training contract numbers, and junior underwriter hiring across Lloyd’s syndicates through 2026 and 2027. If the trend is downward while premium volume holds steady, algorithmic follow is absorbing work that would have gone to entry-level staff.

Why This Matters for Actuaries

The AIG-McGill deal is the clearest signal yet that agentic AI is moving from internal productivity tools to external, market-facing deployment in insurance. For actuaries, this creates both risk and opportunity.

The risk is straightforward: if algorithmic systems can make follow underwriting decisions at scale, the demand for traditional actuarial support of follow portfolios decreases. Loss picks on followed business, reserve analysis for follow syndicates, and rate adequacy reviews for follow capacity all become simpler when the underwriting is performed by a system with continuous portfolio-level awareness.

The opportunity is equally clear: someone has to validate these systems. ASOP No. 56 requires actuaries to assess models used in actuarial work, and an agentic AI system making $1.6 billion in capacity allocation decisions is unambiguously within scope. Actuaries who can evaluate the risk selection criteria embedded in AI follow systems, stress-test portfolio correlation assumptions, and monitor algorithmic performance against loss expectations will be more valuable than those who simply develop loss triangles for the resulting book.

The profession has been discussing AI’s impact in theoretical terms for several years. The AIG-McGill deal, combined with Ki’s demonstrated profitability, Travelers’ 10,000-person deployment, and Chubb’s 20% headcount reduction plan, makes the discussion concrete. Algorithmic underwriting is here, it works, and it is scaling. The actuarial question is no longer whether to engage with it, but how to ensure that the profession remains the authoritative voice on the risk implications of systems that are increasingly making decisions that actuaries used to review.