Various Musings

A couple of thoughts from a week of interesting conversations across my network

This week’s piece is a bit different from others - a lot has happened recently in public & private markets; and the two seem to be more inextricably linked than ever before. I had several meetings with founders and other investors this week that had me thinking about a few different subjects as I got ready to write. So, I thought it might be both timely and a nice change of pace to share a few of those musings in shorter form.

The Recent SaaS Sell-Off

On January 8^th, I wrote about AI and the Next Generation of Systems of Record, in which I highlighted “why systems of record are vulnerable in the AI era.” Since then, SaaS whales like Salesforce, ServiceNow, HubSpot, Workday, and Atlassian have traded down between 25-40%; >$100bn of aggregate Enterprise Value – just from those 5 companies – wiped out in ~30 days. How much of that sell-off was caused by my write-up? Probably less than Anthropic’s launch of plug-ins for Claude Cowork… but we’ll never know for sure. Regardless of the culprit, it’s clear that public market investors are concerned that the AI trade is reshaping markets.

The MAG 7 stocks – Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, Tesla – faired comparatively better over that timeframe; down ~3% in aggregate. Why? Without spending too much time on each company in particular, the key differentiator between the first group of companies above and the MAG 7 is CapEx. In 2025, the MAG 7 had a combined capital expenditure of ~$400bn. The 5 SaaS companies above had an aggregate CapEx of ~$2bn. Put differently: the MAG 7 have the balance sheets, the capability, and the intent to invest heavily into AI as part of their core business models – legacy SaaS do not (as of now).

I ended that write-up saying, “In theory, [legacy SaaS] should be in pole position to win within the AI era, as well. But software margins are exceptionally high – AI margins are considerably lower. Each incumbent’s position of dominance in this new era will be determined by its willingness to shrink gross margins in order to grow gross profits.” In a perverse way, this major sell-off may be the positive signal that legacy SaaS needs to begin investing more aggressively in native AI tools. Public market investors underwrite future cash flows – if they’re predicting diminishing future cash flows for legacy SaaS, it reduces the risk for companies to invest heavily in fundamental business model changes.

I find myself increasingly analyzing public companies like private ones: who has the leadership, product vision, and talent density to compete in the age of AI? The answers here will determine which legacy SaaS companies thrive vs. survive vs. lose in the next ~5 years. Structural thinking - e.g. “Company XYZ is structurally set up to win medium-to-long term because of years of infrastructure positioning” - is less useful in my opinion, as timelines for building software compress rapidly.

Financial Fraud in the Age of Agentic Commerce

For the past decade, fraud instances have scaled alongside payment digitization: as online payment volume, instant settlements, and payments APIs increased, so too did the vectors for fraud. The industry responded with rules engines, heuristics, and eventually machine-learning models trained on historical patterns to combat rising fraud. That arms race largely assumed a human adversary operating at human speed.

As AI agents increasingly initiate transactions, negotiate prices, move money, onboard accounts, and interact with financial infrastructure on behalf of users and businesses, we are introducing a new actor into the system – one that operates continuously, autonomously, and at machine speed. Fraud, predictably, will follow.

Vendors that interact with these AI agents need agent-specific fraud detection systems that scale accordingly; but fraud models are built on historical data. If AI agents are executing hundreds or thousands of payments ahead of fraud models, vendors bear the risk. In these early days of agentic commerce, vendors have chosen to avoid this risk altogether and shut out agents from transacting on their platforms. But this isn’t the solution either – agentic commerce can and will increase vendor efficiency and throughput. Therein lies the dilemma that vendors are faced with, and the opportunity for companies building fraud solutions and agentic payment rails.

Companies that succeed at navigating the new normal of agentic commerce will have fraud systems that:

• Sit early in the decision loop, not after the transaction

• Control permissions in real time, as opposed to just monitoring outcomes

• Understand context around individual payments

• Price risk dynamically as agents evolve

Fraud companies have a big opportunity to serve SMBs in particular. Small businesses don’t have fraud teams, legal departments, or tolerance for ambiguity. As they adopt agentic tools – often bundled into vertical software – they’ll outsource trust by default.

What’s Next within Traditional Financial Services?

This section probably deserves its own extended write-up, so perhaps I’ll revisit it soon. But I’m curious to see what happens within “traditional” financial services in this AI cycle – e.g. payments, lending, mortgage, banking. In previous cycles – internet, mobile, cloud – the fintech asset class thrived, as digital processes made these “traditional” financial services a more delightful process, and a more accessible one.

Take PayJoy, for example – one of my favorite companies from the last cycle. Founded in 2015, they offer point-of-sale cash loans to underbanked consumers in emerging markets who want smartphones. They use alternative data and ML models to underwrite those consumers. They use their patented phone-locking technology to incentivize users to repay their loan when they’re delinquent. As a result, PayJoy is able to provide loans at APRs that are at least 50% less than their closest competitors. Consumers are able to build credit scores from scratch and enter the financial services ecosystem, and they’re able to partake in the digital economy thanks to their smartphones, which in many cases they otherwise could not reasonably afford. In the process of improving consumers’ lives, the company has also done exceedingly well. PayJoy’s model was not possible in a world before internet, mobile, and cloud – now, millions are better off because of how PayJoy combined those technologies.

There are some clear use cases for AI in fintech – I covered fraud & agentic payments above, for example – but we haven’t yet seen how these solutions will trickle down to affect the consumer/SMB banking experience. So far, companies like LDGR Systems (a Shirlawn Capital portfolio company) are automating human tasks in the financial services back-office. Loan Origination Software like Vesta automates aspects of a Loan Officer’s job. Ramp is probably the highest profile case of how AI in fintech can save users time and money.

As these solutions make further inroads within their respective fields, the customer experience will continue to improve – but following Ramp’s lead, we may also find that more money winds up in the end user’s pockets. As these AI solutions reduce operational overhead, companies can pass on part of these savings to consumers in the form of rewards, APYs/APRs, and more. And if legacy companies think they can withhold these savings and keep them for their own bottom line… we’ll see a whole new wave of AI neobanks and other end-to-end AI financial services companies.

We’re still very early in the AI-for-fintech evolution; especially as it pertains to “traditional” financial products.