Margin of Safety #43: SaaSpocalypse, Vibe Coding, and the New Scarcity

When software is free, accountability becomes the moat

Equity markets are currently reeling from what analysts have dubbed the SaaSpocalypse. This month, nearly 300 billion dollars in market value vanished in a single trading day following the release of frontier agentic tools [1]. In light of this chaos and the ongoing news cycle around Moltbook, we think Reddit is an interesting case study. The company just reported a fourth-quarter revenue surge of 70 percent, yet its valuation fluctuates wildly [2] as investors weigh its 100 percent human-created content against the hype around agents and autonomous engineering.

The catalyst for this anxiety is vibe coding. As demonstrated by products like Moltbook or platforms like Bolt.new, vibe coding (in theory) allows users to generate functional applications through high-level natural language specifications. If the cost of a feature-complete application drops toward the price of a few million tokens, the traditional software-as-a-moat strategy evaporates. In practice, we’re hearing a decent amount of evidence that true, hands off the steering wheel vibe coding is more vibe than reality, and that the most successful uses of AI in coding are when it’s supervised by experienced engineers – in such a world, costs won’t become pure tokens. But productivity improvements may well lower them, and perhaps dramatically.

For a social giant like Reddit, the threat of vibe coding is ostensibly existential. If a solo developer can use an LLM to vibe a functional clone of a popular subreddit interface into existence this afternoon, why does Reddit still command a multi-billion dollar valuation? The cold start problem—the difficulty of populating a new network—has historically been Reddit’s greatest protection. However, vibe coding, when paired with AI-generated content, could effectively simulate high-fidelity, context-aware engagement to kickstart new communities. AI makes it easy to mimic specific linguistic tics and cultural norms to create a facade of activity that can lure in real users.

Despite this, we think technical specificity reveals the limits of the vibe. While an agent can generate a React frontend, it cannot yet autonomously architect a system capable of managing Reddit’s current high-concurrency write requirements and international privacy compliance. The recent security debacle of Moltbook, which leaked 1.5 million API keys due to unverified and poorly designed agent-written code, serves as a timely reminder of these limits. And while it’s feasible in abstract to kickstart a social network with bots, replicating something the scale of Reddit would require a tremendous number of tokens. Moltbook once again serves as a demonstration, with its bot-authored content being interesting but ultimately unengaging to human users. At least for now, we believe Reddit has the very real potential to survive and even thrive as a bastion of human(ish) connection.

The threat to B2B SaaS is potentially just as existential. Here, vibe coding is not just a tool for competitors; it is a tool for customers who are tired of the SaaS tax. Versus the scale of Reddit, many SaaS businesses deliver services that are low throughput but higher process complexity. These processes often covering broad, high-stakes workflows—such as global tax compliance or multi-entity ERP—that involve interlocking regulatory logic. This is a similar but different from of Reddit’s complexity moat; it also requires a level of precision and deterministic output that vibing alone cannot reliably hit (at least not yet). That said, many unloved tools suffer from artificial complexity. They feel deep only because of poor user experience and menu-diving friction (at least one of this blogs’ authors has less decidedly than fond memories of old school expense management systems that have been displaced by Expensify et all in the modern startup). The more a service is injecting artificial complexity versus genuinely representing a complex system, the more a small, AI-enabled team can potentially undermine them creating a streamlined, bespoke interface that performs the same function without the overhead of a legacy seat license.

Notably, the small, AI-enabled team doesn’t have to be external. The build versus buy debate is currently being reinvigorated by AI, with enterprise teams increasingly tempted to vibe-code internal replacements for expensive SaaS contracts. But this ignores the total cost of ownership. Software requires ongoing maintenance, and even with vibe coding a human needs to ultimately own that process. Replacing a vendor with an internal tool means assuming responsibility for security patches or vulnerabilities, managing the update or deprecation of third-party services the tool relies on, and ensuring the tool keeps pace with the changing needs of its users. In the context of regulated environments, the needs may not even be fully controlled by the enterprise – they may be dictated by external authorities.

Crucially, vendors provide accountability. A primary feature of an enterprise SaaS contract is the service level agreement. For an executive sponsor, a subscription fee can be an insurance premium. If an internal tool fails, the IT or engineering department is at fault. If a vendor fails, there is an external party to hold liable. In the corporate hierarchy of 2026, the ability to shift blame is a high-value feature that no AI agent can provide.

All that said, we believe a significant threat to SaaS incumbents is the innovator’s dilemma regarding AI integration. As AI agents increase per-user productivity, traditional per-seat SaaS pricing models can be difficult to sustain. If an AI agent allows one employee to do the work of ten, a vendor charging per seat sees its revenue collapse by 90 percent if it successfully helps the customer. It may also see an increase in its own delivery costs, as the agent makes greater user of its service APIs. On the other hand, revamping pricing can be highly disruptive to the existing business (and its sales force). This creates a structural opening for AI-native entrants to embrace outcome-based pricing, aligning their incentives with the customer’s desire for automation. Incumbents, handcuffed by historical revenue targets and investor expectations, are finding this transition nearly impossible to make.

Reddit is not cooked, but it is being forced to reframe its strategy, likely around the scarcity and value of human generated, human curated content and connection. Its future lies in being the authenticated corner of an increasingly synthetic web. For the SaaS world, the secret code era is ending. In a world where basic software (and connectors between enterprise systems) becomes increasingly commoditized, the winning scarcities are human attention and institutional accountability. The winners of 2026 are the orchestrators who realize that while AI can write the code, humans must still write the story and sign the contract. Vibe coding reduces the cost of creation, but it does not change the cost of being wrong. As software becomes abundant, trust re-emerges as the true moat. And trust, unlike code, is very hard to be vibed into existence.

If you’re building something in this space, feel free to reach out to jpark@forgepointcap.com and kshih@forgepointcap.com.

This blog is also published on Margin of Safety, Jimmy and Kathryn’s Substack, as they research the practical sides of security + AI so you don’t have to.