Letter 2 | A New Dawn in the Digital Force:

December 7 , 2025

As we approach the end of the year and I begin writing Why Capital's 2H25 letter, I decided to share one of my main current investment thesis: Software Embedded AI.

I have spent the last five years researching the software industry and studying more than 50 software companies. I have never been more excited about the prospects of investing in software. Many argue that “SaaS is dead” or that AI will end software businesses entirely, which I believe is a superficial interpretation of AI’s impact - not only on software but on the entire economy.

 Just as previous innovations - cloud, mobile, embedded finance, etc - did not wipe out software solutions, I believe AI will not eliminate all players. However, AI will profoundly reshape the software space, just as it is reshaping many industries. What will change is how software companies create value, how they distribute software, how they develop new modules and features, how they compete, and how they monetize the value they deliver.

Since the early days of on-premises solutions, software functioned primarily as a system of record, a place where companies recorded and stored data. With the cloud and mobile revolutions, software became cheaper to distribute, acquire, and use - and its value proposition expanded. In addition to being a system of record, software became a system of interaction, enabling companies to create ecosystems where they interact with suppliers, transact with customers, and allow employees to exchange information with colleagues.

One of the adjacent solutions that expanded software companies’ total addressable markets and increased average tickets was embedded finance. As systems of record evolved into systems of interaction, some software providers embedded payments and other financial services directly into their workflows. This allowed them to move from purely subscription-based SaaS models to transactional-based revenue models, charging customers a fee or a take-rate over the TPV (total payment value) transacted within their ecosystem.

Many listed companies - including AppFolio, Toast, Xero, Intuit, Shopify, ServiceTitan, and Bill.com - have meaningfully grown revenue through this transactional model. Other companies such as Salesforce, Hubspot, Monday, Atlassian, Procore, Twilio, etc have not achieved that to the same extent or at any extent at all.

But what exactly is the Software Embedded AI thesis, and how does it relate to the software Embedded Finance innovation wave?

The Software Embedded AI thesis is basically the belief that certain great software businesses with durable moats and strong competitive advantages are well-positioned to embed AI features and build AI agents within their solutions - just as some successfully did with embedded finance - which will substantially increase the value they deliver to customers while also strengthening their competitive advantages.

It is the system of record and the system of interaction evolving into the system of engagement. By “system of engagement,” I mean that software companies will be able to leverage the advantage of their core products to build powerful AI agents and features that allow them to capture a larger share of their customers’ P&L. As a system of record and interaction, a software provider can only capture a limited percentage of customer revenue. But as these companies start automating tasks previously performed by humans, they can begin to tap into labor payroll costs and expenses.

To give a sense of scale: global payroll costs and expenses are roughly $50 trillion. If software businesses are able to increase employee productivity - or even partially substitute labor - by just 20%, it is not a stretch to say they could capture at least $10 trillion in revenue over time. For perspective, that would be more than 15× the total CapEx hyperscalers are expected to deploy together to build data centers in 2025.

However, in the same way AI represents a massive opportunity to expand software companies’ TAM, it also poses significant threats. Just to mention two risks: (i) developer productivity is accelerating rapidly with AI, reducing the cost of building good software and lowering barriers to entry in the industry and (ii) mature software companies, even those with strong cores and wide moats, may face product disruption if their underlying architecture and products becomes outdated - similar to how cloud-based software disrupted on-premises solutions.

So, how do we identify which software companies are well-positioned to benefit from the AI revolution - and which ones appear at risk? It is a simple question, but the answer is quite tricky. A useful way to think about it is to revisit the Embedded Finance wave, which began to take shape less than a decade ago.

The companies that succeeded in Embedded Finance - such as AppFolio, Toast, Xero, Intuit, and Shopify - were those uniquely positioned at critical workflow or transactional control points within their customers’ businesses. AppFolio became the operating system for property managers, making rental-payment processing a natural extension. Toast’s POS sits at the exact moment restaurants transact, enabling a transaction-based model. ERPs like Xero and Intuit, along with AR/AP automation platforms, manage core financial workflows with deep customer data, giving them an inherent advantage in payments and credit. Shopify, by sitting at the intersection of buyers and sellers, was able to capture a share of the value flowing through its marketplace infrastructure. These examples illustrate that only software companies deeply embedded in customer workflows, data, or transaction control points were positioned to successfully layer new financial solutions and revenue models.

In the Embedded AI era, I believe something similar will happen. Software platforms used by “desk workers” in manual, repetitive workflows and cognitive tasks - such as analysis, scheduling, documenting, selling, and supporting customers - will be able to automate those workflows through AI agents and AI-native features. These solutions will meaningfully increase employee productivity or substitute labor altogether. Because of that, I spend more time studying companies operating in these domains rather than software companies whose users are “hand workers,” such as restaurant attendants, plumbers, electricians, construction workers, and similar roles. The latter will most likely feel the impact of AI primarily through robotics, which still seems to be further out in the future than the former.

For instance, AppFolio launched Realm-X, which sits on top of property data to automate leasing conversations, email drafting, and real-time portfolio insights for property-management teams. Salesforce has its Agentforce 360 suite, which already generates CRM content (emails, summaries, responses), surfaces predictions and recommendations, and runs text- and voice-based conversational AI agents across sales, marketing, commerce, and customer support. Bill uses AI to read invoices and W-9s, extract data, auto-categorize expenses, detect anomalies, and automate AP/AR workflows for both SMBs and accountants. Even EDA software companies are deploying human-substituting tools, as Cadence Design Systems did by developing Cadence.AI - an agentic platform that enables chip designers to innovate faster and shorten system-on-chip (SoC) engineering cycles by months, all while requiring fewer engineers.

However, there are other great software companies that - even though they have strong competitive advantages and successfully executed the embedded-finance play - will likely struggle to access the “payroll money pocket” in the industries they serve. I don’t see Toast substituting most restaurant employees, nor ServiceTitan replacing plumbers and electricians, anytime soon. Of course, these companies are implementing AI solutions in peripheral adjacencies that may add incremental value, but I do not see how they will capture a substantially larger portion of the value chain’s profit pool through AI alone and how they will become an even more mission-critical tools for their customers in the era of AI.

I’m not arguing that AppFolio, Salesforce, Bill or EDA software companies will necessarily be the winners of the AI era, nor that Toast or ServiceTitan will be losers. The key point is that specific players and software categories are inherently better positioned for Embedded AI agent use cases, making first-mover advantages not just possible but likely. These conditions may allow some companies to build new, strong, and wide moats and become even more mission-critical to their customers - if they can harness the opportunities and mitigate the risks faster than others that lag or that will be able to capture less incremental value from AI.

It’s also worth noting that, just as in the cloud era, product quality is critical in software, but it is not the only source of advantage. There are several additional ways to build incremental competitive advantages that allow certain players to sustain high marginal ROICs for years. Distribution channels, brand, switching costs, network effects, data moats, and deep vertical-domain expertise - among others - will remain essential for widening and protecting software companies' moats.