After AI Agents became popular, many people started writing obituaries for SaaS. But I think it's too early.
Investors are indeed very panicked. In early 2026, the panic of the SaaS doomsday swept through the entire tech industry. At the end of January, Anthropic simply released a feature update that allowed Claude to call plugins, and the market value of the US software sector evaporated by hundreds of billions of dollars in the following three weeks.
The logic behind their panic is very simple. They believe that since AI can already write code, find vulnerabilities, and even dynamically generate tools, the cost of writing code is infinitesimally close to zero. Once Agents can create various customized tools for enterprises anytime, anywhere, the moats that those software companies that collect rent monthly will naturally disappear.
Their moats will naturally disappear.
... Thus, from CrowdStrike to IBM, from Salesforce to ServiceNow, no matter how impressive their financial reports were, they were all experiencing brutal sell-offs. Meanwhile, countless AI entrepreneurs were holding business plans, telling VCs they wanted to "build the middle layer for the Agent era" and "start a business for Agents." They were all betting on one thing: making tools was the sexiest business of the era. But if we shift our focus away from those PowerPoint presentations and look at the realities of how businesses operate, we'll find that it's not like that at all. Software is never just about selling code. There's a classic and repeatedly validated theory in economics called "factor scarcity transfer." Every revolution in productivity makes a previously scarce factor abundant, while simultaneously making another previously neglected factor extremely scarce, leading to the concentration of wealth in the latter. Before the Industrial Revolution, labor was scarce; the steam engine made mechanical labor abundant, transferring scarcity to capital and factories, making factory owners the wealthiest people of that era. The internet revolution reduced the cost of information dissemination to zero, shifting scarcity to users' "attention," making traffic a huge business. Today, the AI revolution is making the ability to write code and create tools extremely abundant. In the Agent era, where code is no longer scarce, where has the scarcity shifted? In fact, in the decades of software industry development, code itself has never truly been a moat. Every line of code in the Linux system is free, but that didn't stop Red Hat from being acquired by IBM for a staggering $34 billion; MySQL is free, yet Oracle, after acquiring it, still manages to sell expensive service contracts with it. Anyone can download PostgreSQL's code, but AWS's Aurora database service still rakes in billions of dollars annually from enterprise customers. The code is free, but the business is still there, and doing quite well. The key factors are actually these three things: solidified business processes, years of accumulated customer data, and the resulting extremely high switching costs. When you buy Salesforce, you're not buying the source code of the CRM system, but rather the more than 50 trillion enterprise customer records it manages, and the experience it has in seamlessly integrating sales, customer service, and marketing processes. These data aren't just lines of cold, hard code; they represent a company's living time and history. A company that's used Salesforce for ten years has every customer communication record, every transaction history, and every follow-up point for a sales opportunity all stored there. Migrating isn't just about switching software; it's like moving the entire company's memory. This is why Salesforce still generates $41 billion in annual revenue and has set a $63 billion target for 2030. Returning to the framework of factor scarcity transfer... Since agents can create their own tools, and the cost of writing code is practically zero, what is the most scarce element in the enterprise service scenario? The real bottleneck for agents isn't their lack of tools, but their lack of "context" in their brains. A super agent with all the tools is like a top-of-the-line juicer. It spins incredibly fast, its blades are sharp, but if no one puts fruit in, it certainly can't produce a glass of juice. McKinsey's annual report points out that 88% of enterprises are using AI, but only 23% have truly achieved large-scale deployment of agent systems within some part of their operations. What's holding them back isn't that the large models aren't smart enough, but that the enterprise's data architecture isn't prepared. In an interview with MIT Technology Review, SAP's President of Data and Analytics, Irfan Khan, stated, "Companies can't simply throw away their entire general ledger system and replace it with an agent, because an agent can't do anything without business context." This "business context" refers to: the company's financial compliance baseline, the industry's regulatory requirements, the current customer's preferences and history over the past decade, the supplier's payment terms and default records, the employee's performance history and promotion path… This information is neither publicly available on the internet nor obtainable through web scraping, and it's something AI cannot generate through text prediction. Ashu Garg, a partner at Foundation Capital, shares this view. He says that agents need more than just data; they need a "context graph," a reasoning layer that captures not only what the company has done but also how it thinks. This kind of thing can only be accumulated from real business operations and cannot be created out of thin air. Under this logic, scarcity has shifted from "the ability to create tools" to "possessing irreplaceable business context data." Since an agent can't conjure up a glass of juice on its own, who actually holds the fruit? The Golden Age of Data Landlords The answer points to those old-timers once thought to be disrupted by AI. On February 23, 2026, Bloomberg launched an agentic AI interface called "ASKB." Bloomberg Terminal is one of the most representative examples in the software industry. Although it only has 325,000 subscribers globally, each account charges $32,000 annually. This means that Bloomberg earns over $10 billion annually from these 325,000 accounts, accounting for more than 85% of Bloomberg LP's total revenue. For the internet industry, where "the more users, the better," this is actually illogical. Bloomberg has built a solid commercial fortress with a very small number of paying users. The reason it can succeed is simple: Bloomberg possesses the most complete, real-time, and deeply structured financial data globally. This data is the product of decades of continuous investment, including real-time market data, historical archives, news corpora, analyst reports, and company financial data… Any institution that wants to make serious decisions in the financial field has no choice but to use it. For the newly launched ASKB, AI is the engine, and Bloomberg's unique data is the only fuel. Any agent that wants to play a role in the financial field cannot fabricate this data out of thin air; it can only obediently connect to Bloomberg's interface. WatersTechnology gave a very insightful comment: Bloomberg's agentic approach demonstrates "how those who own the data turn AI into their own ATM." This logic applies to various vertical industries. Veeva possesses compliance and R&D data for the global pharmaceutical industry; any pharmaceutical company's agent must access this data to handle clinical trials and regulatory filings. Epic holds the medical and health records of over 250 million patients in the United States; every diagnostic recommendation from a medical agent requires this real medical record data as a foundation. LexisNexis monopolizes a vast legal document archive; legal agents cannot bypass it for case searches and compliance analysis. This data represents the culmination of decades of real-world business operations, a sedimentation of time, and an irreplaceable history. This is also the ultimate manifestation of the "transfer of factor scarcity": when everyone possesses top-tier AI engines, what truly determines victory is whether you can find your own unique oil field. In the past, these subscription-based data services were sold to human analysts. A large organization might need to purchase 100 Bloomberg terminal accounts. But in the future, when machines become consumers of data, an organization might be running tens of thousands of agents, frantically calling these proprietary data interfaces within milliseconds. This is a leap in scale. A human analyst can only handle a limited number of queries per day, but agents call them far more frequently. The demand for continuous, real-time, high-value data will explode exponentially. The subscription-based business logic hasn't been overturned; instead, it's been amplified infinitely by the machine's insatiable appetite. Code goes to zero, data starts collecting rent. But does this mean all SaaS and data companies can rest easy? Not all SaaS companies have this advantage. If you interpret this article as indiscriminately bullish on the SaaS industry, you're sorely mistaken. AI is bringing a brutal differentiation to SaaS. In early March 2026, TechCrunch interviewed several top VCs, asking them what they least wanted to invest in. Silicon Valley investors are already voting with their feet. Simple workflow encapsulation, cross-industry applicable tools, and lightweight project management—these stories that once fueled a funding round are now being directly rejected. The reason is simple: agents can do these things anytime. Software companies without proprietary data are rapidly losing their chance to attract capital. This assessment has divided the SaaS world in two. One half consists of thinly packaged tool-type products, simply putting a nice-looking interface on public data, or merely optimizing a single point of operation. The moat for these products is essentially user habits and interface stickiness. But as Jake Saper of Emergence Capital said, "Previously, getting humans to form habits in your software was a powerful moat. But if agents are doing these tasks, who cares about human workflows?" This type of SaaS does indeed face a significant threat. The GTM toolchain is a prime example. Gainsight, Zendesk, Outreach, Clari, and Gong each occupy adjacent functions such as customer success, customer service, sales outreach, revenue forecasting, and call analysis, each requiring separate budgets, operations, and integrations. AI-native companies can now use an agent to connect all these links, significantly diminishing the value of these isolated tools. Meanwhile, the other half of SaaS is deeply embedded in core enterprise business processes, possessing irreplaceable proprietary data. These companies will not only not be replaced by agents, but will actually become more valuable because of the existence of agents. Taking Salesforce as an example, in February 2026, Salesforce's financial report showed that Agentforce's annual recurring revenue reached $800 million, a year-on-year increase of 169%; it had delivered a cumulative total of 2.4 billion "Agentic Work Units" and processed nearly 20 trillion tokens; it had signed more than 29,000 Agentforce customers, a quarter-on-quarter increase of 50%. More importantly, the combined ARR of Agentforce and Data 360 exceeded $2.9 billion, a year-on-year increase of more than 200%. Marc Benioff said in the earnings call, "We have rebuilt Salesforce into the operating system of Agentic Enterprise. The more AI replaces jobs, the more valuable Salesforce becomes." Salesforce has not only not been replaced by agents, but has also become the fertile ground for agents to operate. Its value lies precisely in the business data and process context that it possesses, which agents cannot bypass. ServiceNow CEO Bill McDermott publicly announced in February 2026: "We are not a SaaS company." He wasn't denying himself, but actively distancing himself. His logic is that SaaS is a concept about "software delivery methods," while ServiceNow aims to become the orchestration and execution layer of enterprise AI agents. AI can identify problems and provide suggestions, but the actual actions executed within the enterprise system must be carried out by platforms like ServiceNow that are deeply embedded in workflows. Workday launched "Sana" on March 17, 2026, a conversational AI suite that deeply integrates HR and financial data. The core logic of this product is not to replace Workday with AI, but to feed AI with Workday's data. Workday possesses salary, performance, organizational structure, and financial budget data from thousands of companies. The depth and uniqueness of this data are something no AI-native startup can replicate in the short term. Therefore, the real moat is not whether you have data, but whether the data you possess is something others cannot obtain, buy, or create. Who Will Collect Rent in the Next Decade? In every technological revolution, those who ultimately reap the greatest profits are often not the inventors of the groundbreaking new technology, but rather those who quietly master the scarce resources upon which that technology depends. In this era of rapid AI development, the capabilities of large models will become increasingly powerful, and the ability of agents to write their own code and create tools will become increasingly widespread. When these capabilities, once considered cutting-edge technologies, become infrastructure, the logic of "transfer of resource scarcity" boils down to one conclusion: those who frantically build tools for agents are unlikely to be the ultimate winners of this era. Foundation Capital, in its February 2026 analysis, stated that the overall market capitalization of the software industry will expand tenfold in the next decade. However, this tenfold growth will not be evenly distributed among all software companies; it will be highly concentrated in the hands of those players who can truly master the agent era. The real winners are those who hold data assets that agents can't bypass. For today's entrepreneurs and investors, there are only two fates for entrepreneurs in this era: one is to desperately build tools for agents, and the other is to first occupy that piece of land. You should know which one you're doing right now. Don't focus on the agent's hand; try to strangle the agent.