Software Companies Must Become AI Companies, Or Perish

In the near future, you'll be able to rehire the best software engineer you ever worked with at a fraction of the cost.

In the time since you last worked together, the engineer gained superhuman capabilities. They can now accomplish in minutes or hours what used to take them weeks or months. They can read documentation, perform deep dives, and absorb information at lightning speed. They can churn out fully functioning apps with minimal oversight, customized to your exact needs. And once they're done, they will happily retire until the next project.

These engineers won't be sitting next to you at the office—they'll be virtual employees who can read, write, and code at superhuman speed. And this transformation will fundamentally reshape the software industry over the next decade.

Much has been written about how this threatens the SaaS business model. When customers can easily build custom software in-house, they will balk at recurring subscription fees. Large, tech-forward enterprises like Klarna are already bringing more software in-house with the help of AI¹

This poses a fundamental challenge: Since the cost of recreating their products is shrinking rapidly, SaaS companies must reinvent themselves. In this post, I'll argue that their survival depends on pivoting from selling platforms to providing the virtual workers who can operate these systems at a fraction of human cost.

Labor-augmenting vs. labor-replacing

Let's distinguish between two types of software: labor-augmenting tools that make humans more productive, and labor-replacing systems that actually do the work humans once performed.

Most SaaS companies today sell software that is principally labor-augmenting, not labor-replacing. Their products create slick interfaces that let people manage complex systems of record without deep technical knowledge. But humans still bear the operational burden—whether the software is built in-house or purchased off-the-shelf.

This framework helps us understand both the current SaaS landscape and its vulnerability to AI disruption:

	Low complexity	High complexity
Labor-augmenting	Traditional SaaS interfaces and dashboards Examples: - Calendly - Trello - Zendesk	Complex workflow automation systems Examples: - Salesforce - Workday - ServiceNow
Labor-replacing	AI-generated CRUD apps and internal tools	Autonomous agents performing human cognitive work Examples: - Harvey - Devin - AI software engineers
The most vulnerable SaaS products sit in the upper-left quadrant—simple interfaces that merely help humans work more efficiently. These are precisely the products that companies will most easily recreate with AI tools. Products in the upper-right quadrant (complex automation) have a longer runway but will eventually face similar pressure.

The lower quadrants represent the AI-driven future. The lower-left shows what customers are already building in-house, while the lower-right represents the next frontier—and the greatest opportunity for SaaS providers.

Here's the good news for SaaS companies: The true economic value of AI isn't in making it cheaper to rebuild existing software. The benefits of AI come from its ability to do things that weren't possible before. If a thousand companies rebuild Salesforce internally, that adds little to no overall productivity. If a thousand companies each gain a new virtual employee, there will be a thousand more productive companies.

This reveals the path forward for SaaS companies: They must evolve from providing tools that augment human labor to offering virtual workers that augment workforces.

In-house development

As AI drives the cost of software development toward zero, companies will inevitably bring more development in-house. But there's a critical distinction between building software and creating the virtual workers who can operate it.

Let's see how this might play out in practice.

ACME Corp recently deployed AI coding tools to build a custom internal sales portal. The project was a stunning success—the new system not only replicated their previous SaaS platform's functionality but also incorporated features they'd been requesting from vendors for months.

Emboldened by this success, the sales team wanted more. They envisioned virtual coworkers who could pull data, create dashboards, configure workflows, and even handle outbound sales activities.

This is where ACME hit a wall. The AI engineers who easily generated code for their sales portal struggled significantly when tasked with creating autonomous agents. The virtual employees they built were unreliable, error-prone, and incapable of handling complex, multi-step tasks without human intervention. It turns out, agent engineering is an order of magnitude more difficult than code generation.

Agent engineering

While companies can increasingly build their own software, building agents to use that software could remain a challenge. It requires expertise that isn't readily available and sits much further down the complexity tree than simple CRUD applications. This expertise gap means SaaS companies will be able to build robust, generalized agents better than their customers for at least the next few years.

SaaS companies have two advantages here. First, they possess vast troves of user data—events, feedback, and interactions across diverse customers—that capture the cognitive tasks not well-represented in public datasets. Second, they can attract and concentrate AI talent in ways individual customers cannot.

I believe agent-engineering tools won't reach the same level of accessibility as today's no-code web app builders for at least another 1-2 years. This creates a window of opportunity for SaaS providers who act decisively.

A new business model

SaaS companies must transform labor-augmenting platforms into labor-replacing systems. As customers gain the ability to build software in-house, they'll resist traditional subscription fees while eagerly paying for virtual employees that replaces expensive human labor. In this world, the pricing will be based on outcomes rather than usage.

This transition explains the explosion of AI-native vertical SaaS startups. Built from the ground up with this new paradigm, they're positioned to capture emerging markets and disrupt incumbents who fail to evolve quickly enough. For established SaaS providers, the message is clear: Embrace AI, or perish.

---

1. It turns out, most of Klarna's gains were realized from centralizing knowledge in a graph structure, not AI directly. Sebastian Siemiatkowski did mention that AI, when plugged into this graph, became more effective, and that companies need to be AI-first. ↩︎