Did JPMorgan's AI Agents Really Beat the 60/40 Portfolio? And Why the Fed Just Named a16z's Marc Andreessen to Judge AI's Impact on Rates?

Welcome to Agentic Newsletter for Financial Services - the first newsletter dedicated to the transformation in financial services driven by the advancements in AI, shaping the Agentic economy and crafted by a team of more than half a dozen agents that keep evolving and adapting.

These Agents scan the world on a daily basis, filtering hundreds of news articles. We work together to bring you a weekly curated selection coupled with my human insights.

Thanks for reading and being part of our fast-growing subscription community. This is edition No. 57, one year in and past 40,000 subscribers. Let's stay up to date and build a more human world powered by agentic AI.

Week in Review

While fintech's attention this week is fixed on Stripe and Advent International's unsolicited $53 billion offer to take PayPal private, a 28% premium backed by roughly $50 billion in committed bank financing and one of the largest leveraged buyouts in fintech history, the agentic AI story in financial services remains active and lively. 

JPMorgan disclosed that eight AI agents it built on OpenAI and Anthropic models outperformed a traditional 60/40 portfolio across twenty years of backtesting, the best by 70 basis points of annualized return at lower volatility. The bank calls these backtests, not live results, and CEO Jamie Dimon voiced caution about AI reliability the same quarter the bank posted record earnings. A KPMG survey cited by Reuters found 51% of banks already piloting AI agents across wealth, trading, treasury and client vetting, with Morgan Stanley's wealth AI lead now sorting roles into human, hybrid and fully agentic categories.

Mastercard completed its first agentic payments pilot in Moldova, built on a Know-Your-Agent identity framework, continuing a pattern of testing trust architecture in smaller markets, Thailand, Finland, now Moldova, before any G7 rollout. Lyzr let its own AI agent run investor relations for a $100 million Series B, and Monzo co-founder Tom Blomfield left Y Combinator for Anthropic's compute team the same month OpenAI posted a $205,000 role for an investment-banking subject-matter expert, both hires reading as a race for credibility with regulated financial buyers rather than a research hire.

And the Federal Reserve named Andreessen Horowitz co-founder Marc Andreessen to co-lead its new task force on AI, productivity and jobs, an appointment Fed Chair Kevin Warsh compared to Alan Greenspan's 1990s productivity bet on the internet. Andreessen Horowitz manages roughly $90 billion in assets, including billions raised in AI-specific funds in the months before this appointment, and unlike advisory panels at the FDA or the Department of Energy, the Fed is not bound by the Federal Advisory Committee Act's disclosure and recusal requirements.

The week's signal: capital is chasing certainty in the backtests and the pilots, while the people judging whether AI is good for the economy increasingly have a stake in the answer.

1. DEPLOYMENT & USE CASES

1. JPMorgan's AI Agents Beat the 60/40 Portfolio in Two Decades of Backtests

JPMorgan built eight AI agents on OpenAI and Anthropic models, each classifying markets into one of four regimes and reallocating capital accordingly, and every one outperformed a traditional 60/40 stock-bond portfolio on a risk-adjusted basis across roughly twenty years of historical simulation, the best adding 70 basis points of annualized return at lower volatility.

JPMorgan is explicit that these are backtests, not live trading results, a distinction doing most of the work here.

• The bank running trillions in client assets just tested whether an algorithm should set asset allocation, and got numbers good enough to publish.
• CEO Jamie Dimon voiced caution about AI reliability the same week the bank reported a record quarter.

Backtested outperformance routinely fails to survive real transaction costs and live markets, so treat the 70 basis points as a hypothesis worth testing live, not a track record.

2. Half of Wall Street Is Already Piloting AI Agents, and the Roles Are Being Redrawn

A KPMG survey cited by Reuters found 51% of banks piloting AI agents across wealth, trading, treasury and client vetting, with Goldman Sachs, Citi and Morgan Stanley naming concrete deployments and Morgan Stanley's wealth AI lead now defining which roles are human, hybrid or fully agentic.

The 51% isn't the real headline; the redrawing of who does what between people and agents is.

• KPMG's own consulting bias means self-reported pilot numbers likely run ahead of production reality.
• Sorting roles into human, hybrid and agentic categories raises labor and accountability questions no board has resolved yet.

Ask any vendor citing a 51% adoption stat which of those pilots have actually moved past a proof of concept.

2. AGENTIC COMMERCE, RAILS & INFRASTRUCTURE

3. Moldova Just Ran the Agentic Payments Trust Test That Actually Matters

Mastercard completed its first agentic payments pilot with two Moldovan banks, maib and Moldindconbank, built on a Know-Your-Agent framework that requires AI agents to be identified, registered and authenticated before they transact, using agentic tokens, spending limits and revocable consent.

The story isn't that agentic payments work. The story is the pattern: Mastercard keeps proving this architecture in smaller, faster-moving markets, Thailand, Finland, now Moldova, rather than leading with the US or UK, where smaller regulators and smaller bank counterparts let it iterate at lower stakes.

• Watch which market Mastercard picks next; that choice signals how close the architecture is to being considered production-ready at scale.
• Nothing here has been tested against dispute volume or fraud attempts at consumer scale yet, only pilot-stage transaction flows.

3. MARKET IMPACT, FUNDING & INNOVATION

4. A Startup Let Its Own AI Agent Run Its $100 Million Fundraise, and That's the Real Pitch

Lyzr, which builds AI agents for enterprises, used its own agent, SivaClaw, to field investor questions, draft memos and track engagement across a $100 million Series B at a roughly $500 million valuation, with the company reporting $400 million in investor interest without a founder traveling for in-person meetings.

An agent running fundraising logistics is not the same as an agent making financial decisions with real risk attached.

• The $400 million interest figure is self-reported and unverified, and deserves the same skepticism as any unaudited claim.
• Lyzr sold proof-of-concept and product pitch in the same motion, and investors bought both.

Treat the fundraise itself as the case study; the interest figure is marketing until an independent source confirms it.

5. AI Labs Are Recruiting Finance's Own People to Build the Case That They Belong Inside Finance

Monzo and GoCardless co-founder Tom Blomfield left Y Combinator to join Anthropic's compute team, while OpenAI separately posted a $205,000 investment-banking subject-matter-expert role tasked with defining where AI should automate analyst-to-director work versus leave it to human judgment.

Neither hire is about research capability; both read as a bid for credibility with financial-services buyers.

• Anthropic gains an operator who spent a decade building trust into a regulated consumer bank.
• OpenAI gains someone who has lived the investment-banking career ladder from the inside.

Watch-out: the framing that either company specifically wants regulated-sector credibility is analytical inference from the coverage, not a claim either firm made directly.

4. REGULATION & GOVERNANCE

6. The Fed Just Put a16z's Marc Andreessen in Charge of Judging What AI Means for Interest Rates

The Federal Reserve named Andreessen Horowitz co-founder Marc Andreessen to co-lead its new Productivity and Jobs task force, assessing how AI could reshape inflation, growth and monetary policy, an appointment Fed Chair Kevin Warsh compared explicitly to Alan Greenspan's 1990s productivity bet on the internet.

This isn't adjacent to interest rate policy. Productivity assessments feed directly into the Fed's reaction function: if AI is genuinely lifting output without fueling inflation, that's an argument for holding rates lower for longer, and every task force member so far has publicly praised AI's economic upside.

• Andreessen Horowitz manages roughly $90 billion in assets, including $3.4 billion in AI-specific investments raised just six months before this appointment.
• The Fed is exempt from the Federal Advisory Committee Act, the law that requires advisory-committee members at the FDA or the Department of Energy to file financial disclosures and recuse from matters where they hold a stake.

Recommendations aren't due until year-end, which limits near-term impact but doesn't answer whose analytical frame is shaping the input in the meantime.

This newsletter is a human-AI collaboration involving seven AI agents orchestrated by Nicolas C. and my own curation and refinement.