The two largest frontier labs both picked finance and both went live in the same week. That is the headline. The interesting part is that they picked exactly opposite distribution strategies.

Anthropic released ten preconfigured AI agents for the financial sector on May 5, covering investment banks, asset managers, and insurers. OpenAI announced a partnership with PwC the same day, building agents around the "core operating rhythms" of CFO and treasury functions. Anthropic shipped product. OpenAI shipped through a Big Four advisory firm. Same vertical, same week, two opposite plays.

The frontier labs have stopped arguing about whose model is better. They are arguing about who controls the distribution path into regulated industries.

What Anthropic shipped

Anthropic's release covers ten preconfigured templates for the financial services and insurance sector. Investment banking workflows include pitchbook generation, comparable-company analysis, and document review. Asset management templates cover research synthesis, portfolio commentary, and client reporting. Insurance templates target underwriting and claims triage.

The "preconfigured" framing is the tell. These are not platform tools. They are products. A buy-side firm signs a contract, plugs the templates into its existing data and identity infrastructure, and runs them. The integration is engineering work, but the agent itself is finished.

This is the same playbook Anthropic used with Claude Marketplace for the broader enterprise AI OS positioning. Ship the productized layer; let the systems integrators handle deployment. The IPO-ready revenue framing in the coverage is not coincidental. Anthropic is reportedly preparing for a public offering, and a productized vertical suite reads as enterprise software revenue, not API consumption revenue.

The vertical choice matters. Investment banking, asset management, and insurance are three of the most heavily regulated, document-heavy, fee-rich industries in the developed economy. They were always next, after healthcare.

What OpenAI shipped

OpenAI's approach is structurally different. The PwC partnership routes ChatGPT Enterprise and the OpenAI Agent Builder through PwC's consulting practice. PwC clients with active CFO and treasury engagements receive customized AI agents built on top of OpenAI's stack and configured to the client's own workflow. PwC is the system integrator, the trusted advisor, and the change-management partner. OpenAI is the model and the platform.

The framing in the announcement is about "core operating rhythms." Month-end close. Forecast variance analysis. Treasury cash positioning. Audit trail reconstruction. These are not the showcase use cases that demo well at a conference. They are the workflows where a Big Four partner already sits across from a CFO every quarter.

This is the distribution insight. Internal IT at a Fortune 500 financial institution would take twelve to eighteen months to evaluate, integrate, and deploy a frontier AI agent suite. PwC can compress that timeline because it is already inside the procurement and compliance cycles of those clients. The agent does not need to be sold to the CIO. It is sold to the CFO who already pays PwC to redesign the close process.

OpenAI did not productize a finance vertical. It bought the distribution path.

The pattern: two opposite plays, same target

Anthropic ships product and trusts the buyer to integrate. OpenAI ships through the buyer's existing trusted advisor.

Both bypass internal IT. Both target the budget owner directly. Both compress the procurement cycle below the 12-to-18 month enterprise norm. The difference is which side of the deployment friction each lab is willing to absorb.

Anthropic's bet is that a productized agent with a clear scope is easier to buy than a custom build. The buyer trades flexibility for time to value. The integration is real engineering work, but the destination is known.

OpenAI's bet is that the Big Four already has the buyer's trust, the buyer's data access agreements, and the buyer's procurement signature. A frontier model wrapped inside a PwC engagement looks like a PwC engagement to the procurement team, not a vendor evaluation. The friction shifts from selling the technology to scoping the engagement.

The two strategies are not competing for the same buyer. They are competing for different procurement paths into the same industry.

Why finance was always next

Five vertical agentic launches landed in 48 hours on April 22 and 23. Ballerine for merchant fraud. Aurionpro for trade finance. Savvy Wealth for RIA advisors. ChatGPT for Clinicians. Moomoo for retail brokerage. We called the pattern at the time: vertical specialization is no longer a differentiator. It is the baseline.

The frontier-lab playbook for vertical specialization is now legible. Pick a vertical with regulated workflows and high stakes. Ship a specialist product or partnership at the individual professional or firm level. Build the enterprise upsell path. Minimize regulatory friction at launch with a positioning that does not require fresh regulatory clearance. Healthcare went first because OpenAI shipped ChatGPT for Clinicians on April 23. Finance is next because the document density, the fee pool, and the procurement infrastructure are all unusually favorable.

The demand-side data supports the move. The Visa B2AI study, conducted with Morning Consult across 2,512 US respondents in early 2026, found that 53 percent of US businesses would let AI agents negotiate on their behalf, with 88 percent willing to share pricing and inventory data with AI. Both sides of the transaction are ready in B2B. We have argued for the past 12 months in our own State of the Stack research that B2B would scale before consumer agentic commerce did, and that finance would be one of the fastest-moving verticals once it started. May 5 is the day that prediction stopped being a forecast.

Who is exposed

The thin-wrapper finance AI startups. Hebbia, Rogo, AlphaSense and the cohort that built ChatGPT-style interfaces over financial documents now compete with the foundation model itself shipping ten finished agents covering the same workflows. Differentiation through better fine-tuning is still possible, but the floor on what a foundation model alone can do just rose.

The mid-market specialty vendors. Bloomberg, FactSet, S&P Global, and Moody's hold defensible positions through proprietary data, regulator-side reporting infrastructure, and decades of buy-side integration. Foundation-model agents can summarize and analyze, but they cannot replicate the data feeds. The exposure is at the analysis layer, not the data layer.

The Big Four advisory firms not in the deal. Deloitte, EY, and KPMG now compete with PwC on a partnership that fundamentally changes the consulting margin structure. If PwC delivers a treasury redesign with OpenAI agents handling the analyst-level work, the engagement margin improves while the price to the client either drops or stays flat. The other three firms will need their own frontier-lab partnership inside two quarters or face procurement losses on the next renewal cycle.

The investment banks whose internal teams just got skipped. Goldman, Morgan Stanley, JPMorgan, and the bulge-bracket peers have all built internal AI initiatives. Anthropic's preconfigured agents and OpenAI's PwC partnership both compress the time from idea to deployment in ways that internal teams cannot match unless they are given the same model access and the same productization mandate. The internal-versus-external decision becomes urgent.

The unresolved question

Both products operate inside a regulatory frame that has not been written. Investment banks face SEC, FINRA, and OCC supervision. Insurers face state regulators and the NAIC. Asset managers face the SEC and FINRA. None of these bodies have published guidance on agent-initiated financial actions. None have addressed who is liable when an agent's recommendation triggers a regulatory breach. None have clarified whether existing best-execution and fiduciary-duty rules apply to agent-mediated decisions.

The healthcare playbook is the template. The American Medical Association sent letters to three Congressional AI caucuses on April 22, one day before OpenAI shipped ChatGPT for Clinicians. The AMA framed safeguards including FDA review for AI in diagnosis and a prohibition on AI in mental health diagnosis. A vertical lobby moved before the financial regulators did.

The financial-services equivalent has not yet acted. The American Bankers Association, the Securities Industry and Financial Markets Association, and the Insured Retirement Institute all have policy infrastructure capable of moving quickly. None has issued guidance on agent-initiated financial actions to date. The May 5 launches put that silence under fresh pressure.

What to watch

Three signals over the next 90 days.

Who builds the dispute resolution layer for agent-initiated trades. Existing trade-error resolution processes assume a human trader. Agent-initiated trades that breach a fiduciary duty or a best-execution standard need a different infrastructure. Whichever lab, exchange, or custodian ships that layer first sets the framework.

Whether the SEC clarifies how existing rules apply. The SEC has not yet issued staff guidance on agent-initiated investment recommendations. A no-action letter or a sweep examination would be the leading indicator that supervision is moving from observation to enforcement.

Which Big Four firm matches OpenAI-PwC. Deloitte with Anthropic. EY with Google. KPMG with Microsoft. The firm that does not match within 90 days starts losing engagement renewals to the firm that did.

The infrastructure for agentic finance is shipping. The governance for agentic finance is not. That is the same gap we have been mapping for 15 months in State of the Stack, now landing in the largest, most regulated, most fee-rich industry in the developed economy.

Sources

When the agent recommends an asset allocation that loses money, is the fiduciary duty held by the AI lab, the consulting firm that deployed it, or the asset manager that ratified the recommendation?

Charlie Major is a Product Development Manager at Mastercard. The views and opinions expressed in Major Matters are his own and do not represent those of Mastercard.