The Industrialization Era of Software

Anthropic Sales Organization IT Architecture, SaaStr Live Shot

Anthropic Sales Organization IT Architecture, SaaStr Live Shot

The “SaaS Is Dead” Sign Circling the SaaStr Venue

The “SaaS Is Dead” Sign Circling the SaaStr Venue

Over the past two weeks, we conducted an in-depth study tour in the San Francisco Bay Area, attending over 30 industry exchange events and deeply participating in the flagship SaaS event in North America — the SaaStr Annual conference. This tour made it clear to me that the software industry may soon undergo a massive transformation.

Today’s software landscape sees companies directly purchasing a large number of applications. According to Okta Business at Work 2025, the average company runs 101 applications in 2025, a 9% increase year-over-year, and large enterprises (2000+ employees) average as many as 247 applications. According to Zylo 2025 SaaS Management Index, including “shadow IT” where employees purchase software via reimbursement, the average application portfolio per company in 2024 was as high as 275. Employees have to use multiple pieces of software simultaneously; Gartner reports that knowledge workers used an average of 11 applications in 2023, with 5% of employees using 26 or more applications at work.

But the future software landscape will likely resemble an e-commerce ecosystem: a few Agent entrances act as e-commerce platforms, while other software becomes merchants on those platforms. Or it may resemble manufacturing, where a few terminal products directly face consumers (e.g., cars, home appliances), backed by a massive and complex supply chain.

Many people envision a scenario where every individual and organization can use AI to quickly and cheaply develop personalized software that best meets their needs, turning the software industry into a construction crew of numerous coding agents managed by humans.

But I believe this scenario will not be mainstream. The mainstream will be super-agent entrances coupled with complex supply chains, or super e-commerce platforms with a large number of merchants. The Agent ecosystem’s MCP, SKILL, CLI, plugins, etc., together form the prototype of this picture.

I. Super Agent Entrance and “Headless Merchants”

The opportunities for Agent entrances are certainly limited.

Before this trip to the US, we thought vertical-specific Agents were the mainstream narrative in Silicon Valley, given the heavy investment in such companies, e.g., Harvey for law. After arriving, we found that Silicon Valley investors are already doubting the prospects of verticals.

For example, although Harvey has impressive revenue, Anthropic released a Legal solution last week. Some investors believe that a general-purpose Agent connected to legal data leaves Harvey with a rather low moat.

During the SaaStr expo, GTM verticals were the most abundant, with many AI CRMs. Next to Salesforce were Lightfield and then HappyFox. I went to see each demo individually. Lightfield looked like an Agent that understands the CRM model; its operation process seemed no different from Cowork. I asked where their data came from, and they said you can import from Salesforce. They claim to be the fastest-growing AI CRM, with a booth larger than Salesforce’s. But what moat does such a vertical have? HappyFox? Similar, and halfway through the demo, the presenter’s laptop died.

One Chinese team left a strong impression on me. They built a vertical Agent for marketing, founded in 2023 and doing well initially, but this year they felt joint pressure from Cowork, Lobster, and coding agents, realizing they had to pivot — but they hadn’t figured out how.

I think if a vertical’s core is an Agent, it will be very difficult to withstand the impact of general-purpose Agents.

Ultimately, each person and each company will likely use very few Agent entrances, perhaps only one.

As super-agents become the main user entrance, existing software will no longer face users directly but will become “headless” services, part of the Agent supply chain. During this tour, many companies said they simply registered and enabled API keys for software like Linear, then never looked again, operating everything through Claude Code.

This is not just an observation from offline discussions. The a16z article “Open agentic commerce and the end of ads” proposes that future business software will evolve into “Headless Merchants,” giving rise to a new “intent economy.” Since the buyer and user of services become agents, traditional SaaS front-end interfaces, sales teams, and cumbersome subscription checkout processes will become redundant. Software only needs to expose stable API endpoints; agents will autonomously evaluate and invoke them with tasks and budgets.

The first image at the beginning of this article is a live shot from SaaStr of Anthropic’s presentation. Although Anthropic’s sales team still uses Salesforce, Gong, Gmail, etc., they have all been “beheaded” by Claude.

II. The History and Future of Software Supply Chain

In manufacturing, cars and phones are end-customer products, but a car contains tens of thousands of components. Without a supply chain, no matter how great Steve Jobs or Elon Musk were, they couldn’t have created the iPhone or Tesla. Compared to manufacturing, the current supply chain for business software is too short.

During this trip, I visited the Computer History Museum in Silicon Valley and discovered that the legendary SAGE system was actually a very complex system encompassing software, hardware, buildings, and people. This shows how costly it was to build complex software early on, because software had almost no supply chain then; everything had to be built in-house.

Looking back, open source and commercial software have significant differences in supply chain depth.

Platforms like SourceForge, GitHub, and package management tools like apt, npm, pip have created a thriving open source software supply chain. The open source supply chain is very “deep” — installing a package often pulls in a cascade of dependencies. In contrast, the commercial software supply chain has remained very “shallow” to date. Salesforce Exchange and cloud vendors’ marketplaces have always had low revenue shares. The so-called 300 SaaS unicorns in the US almost all face end customers directly. Although many products in the US and Europe offer open APIs that can call each other, this is not a true supply chain, because these products still basically need to acquire end customers themselves; the customer’s IT department does the assembly, rather than downstream vendors assembling upstream components and packaging them for the customer.

Why is the commercial software supply chain currently so “shallow”? The core reason is transaction costs. Open source software has very low transaction costs (only search and evaluation costs, no purchasing cost), while traditional commercial software’s high purchasing communication and checkout friction blocks the possibility of service-to-service nesting.

So, in the Agent era, will the commercial software supply chain remain as shallow as an e-commerce platform (only platform and merchants) or become as deep as the auto industry (multiple tiers of suppliers nested layers deep)?

I believe it will gradually move toward the latter, because the barrier of high transaction costs is likely to be broken. The a16z article points out that agents will use open protocols like x402 combined with stablecoins to execute ultra-low-friction machine-to-machine micropayments. When the settlement cost per API call approaches zero, commercial software will have the foundation for a deeply nested supply chain like manufacturing. Coincidentally, during our Silicon Valley trip, we happened to meet a friend who is building a startup for small payments aimed at agents, and even based on fiat currency.

A deeper supply chain will inevitably lead to a complete deconstruction of current commercial software forms. Many current commercial software packages are large and bloated (e.g., Salesforce’s extremely complex product lines) for two reasons: first, high friction in commercial procurement encourages vendors to bundle everything into a large suite; second, it is costly for human end-users to switch and combine multiple different software packages. In the Agent era, both obstacles disappear.

On one hand, new payment methods will significantly reduce transaction costs between software components. Commercial software will no longer need to be sold as bundles. Like open source software, they will become highly vertical, functionally focused single-point packages due to low invocation costs. On the other hand, while using multiple software simultaneously is painful for humans, agents can efficiently manage and combine a large number of different software components autonomously. This makes it hard for vendors to sell suites, because no single vendor can excel in every category. Together, these two logics will lead to a more discrete and granular commercial software supply chain, with fewer all-in-one “family bucket” software packages and more extremely focused single-point components.

Once the friction of underlying invocation and combination is eliminated, the ecosystem of single-point component supply for agents will thrive. In fact, such an ecosystem prototype is already in motion. For example, during this tour, two companies (TinyFish and Exa) specialize in providing search for agents, offering structured search APIs or real-time data without any search frontend. Another example is Anthropic, which has “beheaded” the front-end UIs of software like Kong and Salesforce, but rather than replacing them, directly calls them as underlying capabilities. Obsidian and Slack are also becoming increasingly popular because they are easily operable by machines in the Agent environment. They have all become key components in this new supply chain.

This is the true beginning of the industrialization era of software. I believe this is a better era for the software industry.

In the highly mature automotive industry, the revenue scale of vehicle manufacturing and the underlying component supply chain is almost 6:4. The parts and components market accounts for half of the trillions of dollars in the entire automotive industry. Top Tier 1 suppliers (e.g., Bosch, Denso) generate tens of billions of dollars in annual revenue. For entrepreneurs, this is a historic opportunity: participate in the supply chain oriented toward agents and strive to become Tier 1 suppliers.

During this visit by Trump, a highly metaphorical seating arrangement appeared at the state banquet: Cao Hui, head of Fuyao Glass (a core supplier for Tesla), sat directly next to “big customer” Elon Musk. And Zhou Qunfei, head of Lens Technology (a core supplier for Apple’s supply chain and Tesla’s robots), also sat at the same table with Musk, Cook, and other bigwigs. These companies, which do not directly face end consumers, tell us: supply chain is as important as the final product.