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

On March 16, 2026, AIG and McGill and Partners announced a collaboration committing AIG capacity across up to $1.6 billion of McGill's specialty gross premiums written, at 25%, allocated to individual risks by agentic AI running on Palantir's Foundry platform.

The distinguishing feature is not the size. It is that the decisions being automated are follow underwriting decisions, taken on live specialty risks in the London subscription market rather than inside AIG's own submission pipeline.

Key Takeaways

• 25% of capacity across up to $1.6 billion of specialty GWP is allocated through McGill's digital broking platform, with AIG's underwriting criteria embedded in the platform rather than applied by an underwriter at the point of decision.
• A lead underwriter typically signs 15% to 20% of a slip and sets the price every follower accepts, so follow underwriting is the part of the market where the analytical burden was always lowest and the automation case strongest.
• Ki Insurance wrote $1.11 billion of gross managed premium in 2025, up 6.9%, with $171.4 million of profit before tax in its first year standalone, which is the evidence that algorithmic following is profitable rather than theoretical.
• AIG processed over 370,000 submissions by end-2025, 74% of the 500,000 target it had set for 2030, and Lexington's middle market property submit-to-bind ratio rose 35%.
• Over $1 billion invested in foundational data technologies over five years is the part competitors cannot buy off the shelf; the foundation models themselves are commercially available.

What Is Being Automated

The deal automates a specific role in a three-century-old workflow, and which role it is determines how much of the market it touches.

A broker prepares a slip summarizing the risk, terms, premium and limit, then takes it to a lead underwriter who negotiates terms, sets the rate and signs for typically 15% to 20% of the risk. The lead does the deep analysis: reviewing submission data, challenging exposure detail, negotiating exclusions. Follow underwriters then subscribe to the remaining capacity at the terms the lead set, accepting the lead's pricing in exchange for lower acquisition cost and a lighter analytical burden.

That second role is what the AIG-McGill system performs. AIG developed underwriting criteria embedded in McGill's platform so the system can evaluate whether a risk fits appetite and allocate capacity without routing each decision to a human. The Foundry ontology maps every entity, risk, exposure and relationship in McGill's portfolio into a structured representation the agents query, giving near real-time exposure data, limit deployment, modeled risk outputs and loss information.

Peter Zaffino called it a "significant opportunity to deliver greater efficiency to the subscription market." Steve McGill was more direct: "This collaboration has the potential to disrupt the dynamics of the subscription market."

The Follow Book Was Always the Cheap Book

The economics work because algorithmic following attacks the cost line that was already the thinnest, and the precedent for the loss line already exists.

Ki Insurance, launched in May 2020 by Brit and Google Cloud as the first algorithmic follow-only Lloyd's syndicate, posted $1.11 billion of gross managed premium in 2025, up 6.9%, with $171.4 million of profit before tax in its first year as a standalone entity. Capacity partners include Beazley, QBE, Aspen and Travelers, with TMK joining in April 2026. That is the demonstration that an algorithm can follow underwrite at an acceptable loss ratio.

AIG-McGill scales it on both axes. The $1.6 billion of covered premium exceeds Ki's roughly $1.1 billion book, and it represents only 25% of AIG's capacity through one broker relationship. The architecture is heavier too: where Ki evaluates risks on Google Cloud infrastructure, the Foundry ontology lets AIG's agents check a new submission against portfolio-level correlation with the other 75% of its book managed through other channels, continuously rather than at quarterly exposure review.

The expense consequence is the immediate one. Follow underwriting has always run cheaper than lead underwriting because the analysis is lighter; running it algorithmically takes the underwriter cost out almost entirely. On $1.6 billion of follow capacity placed with minimal human involvement, the expense ratio advantage can be taken as margin or spent on price, and a competitor staffing the same book conventionally has to choose between matching the price and matching the margin.

The infrastructure behind it is the harder thing to copy. AIG has spent approximately $300 million on data, digital workflow, AI and talent over two years and over $1 billion on foundational data technologies over five, and proved the architecture at Syndicate 2479, the $300 million Lloyd's vehicle launched with Amwins and Blackstone in December 2025 that queries over four million industry data points through Foundry.

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)

What Gets Left in the Human Pool

The complication is that automating the easy half of a two-sided market changes what the other half is holding.

Speed is the first-order effect and it cuts both ways. In a traditional placement there is an information gradient between a lead who has spent days on the risk and follows who see the slip days later, and adverse selection lives in that gap. Minutes-to-bind narrows it, which should improve the quality of what gets followed. The corollary is that risks the algorithm consistently declines do not stop existing; they concentrate in the residual pool that human follow underwriters are still working, at terms a lead set for a market that included the algorithmic capacity.

Reserving inherits a segmentation problem from the same source. Algorithmically selected and traditionally selected risks on the same class of business carry identical coverage and identical pricing terms, because the follow market's whole premise is that everyone signs the lead's slip. They may nonetheless develop differently, and there is no coverage or rating distinction in the data to separate them on. Building separate triangles requires a flag that reflects how a line was signed rather than what was signed, which is not a field the traditional bordereau carries.

Accountability is the third and the least settled. The traditional market assigns responsibility for a follow line to the underwriter who signed it. When an agentic system commits AIG capacity against a Foundry ontology and the risk produces a large loss, the underwriting decision has no signer in that sense. Other syndicates on the same slip may not know that AIG's participation was algorithmic, which matters because a follower's willingness to take a line has always carried information about the risk. As algorithmic follow capacity grows from Ki's $1.11 billion and this deal's $1.6 billion toward a larger share of the market, that signal degrades for everyone reading it.