The Disposable Software Era

Anthropic shipped Cowork in about 10 days using Claude Code itself. Cursor’s AI agents built a browser - 3 million lines of code - in a week with no human intervention. These aren’t startups trying to prove a point. These are the companies building the tools.

What happens to SaaS when building is this cheap?

The Economics Flipped

Software used to be expensive to build and cheap to distribute. That’s the entire SaaS model: amortize high development costs across thousands of customers, deliver via browser, collect monthly fees. The math worked because building was hard.

Now building is fast. Anthropic reports 70-90% of code at the company is written by AI. Claude Code grew from research preview to billion-dollar product in six months. Cursor hit $100M ARR faster than any company in history - 12 months.

The economics are inverting. When building is cheap, “buying” loses its appeal. Why pay $50/seat/month for a tool that does 80% of what you need when you can build the exact thing you need in a weekend?

a16z calls this “disposable software” - applications that never justified engineering investment now make sense because the investment required has collapsed. The constraint is no longer ROI. It’s imagination.

What This Looks Like

The Cowork example is instructive. Four engineers, 10 days, using Claude Code itself. That’s not a prototype. That’s a product.

Cursor’s browser experiment is even more striking - they orchestrated hundreds of AI agents (planners, workers, judges) to produce a Rust-based rendering engine with a custom JavaScript VM. No human wrote any of it. The result “only kind of works” - developers who tried it found an 88% build failure rate and described the codebase as bloated. But that’s almost beside the point. The point is that 3 million lines of code - work that would have required a large team and years - was produced in days. Quality aside, the velocity is real.

This isn’t hypothetical. Last year I built a 200,000-line TypeScript MVP for a startup client in 7 weeks - by myself. Before AI tools, that’s 12-18 months of work for a software team. And that was before the latest model improvements and tooling refinements. The floor keeps rising.

The Opposition Steelmanned

Before declaring SaaS dead, the counterarguments deserve serious treatment.

Security is a real problem. Veracode tested 100+ LLMs and found 45% of code samples failed security tests, introducing OWASP Top 10 vulnerabilities. Java was worst at 72% failure. The Cloud Security Alliance found 62% of AI-generated code contains design flaws or known vulnerabilities. CodeRabbit’s analysis showed AI code has 1.75x more logic errors, 1.57x more security findings, and 2.74x more XSS vulnerabilities than human-written code.

The “vibe coding hangover” is real. Speed up front, chaos downstream.

Enterprise systems have moats. Compliance frameworks like GDPR and SOC 2 are hard to replicate. Systems of record - clinical trial management, financial reconciliation, regulated industries - have data moats and regulatory logic that custom solutions can’t easily absorb.

Bain’s analysis is useful here: workflows that rely on human judgment, proprietary data, and regulatory oversight are defensible. The CRM you built in a weekend won’t replace Salesforce for companies that need 15 years of institutional knowledge encoded in their data model.

History suggests transitions expand rather than replace. “Cloud will replace on-prem” became “hybrid is the norm.” Global SaaS is still growing - $197B in 2023, $232B projected for 2025. These transitions are slower and messier than they look in the moment.

The 80/20 problem. AI gets you 80% of the way fast, but the last 20% - reliable, secure, maintainable - still requires engineering judgment. Getting to “works on my machine” is trivial. Getting to “runs in production under load without leaking data” is not.

The Synthesis

SaaS doesn’t die overnight. It evolves.

Seat-based pricing makes less sense when the software required to service a seat can be generated on demand. Outcome-based pricing rises. The winning position for existing SaaS players is probably unique data plus AI augmentation - hard to replicate datasets wrapped in AI-powered interfaces.

For most businesses, the calculation is shifting. Custom solutions that were never worth the engineering investment are now worth considering. Not for everything - you’re not going to build your own Stripe - but for the long tail of internal tools, specific workflows, and niche applications.

The Uncertain Ground

The question isn’t whether software development economics are changing. The examples are too stark to ignore.

The question is how fast.

For SaaS companies: adapt or become the next generation’s legacy system. The seat-based model is under pressure. The companies with unique data and real switching costs will survive. The ones competing on features alone are vulnerable.

For businesses: the math is starting to favor custom solutions in places it never did before. Worth running the numbers. But don’t assume AI-generated code is production-ready without the same scrutiny you’d apply to any new vendor.

No confident predictions here. Just a recognition that the ground is shifting - and that uncertainty should inform strategy without paralyzing it.