Your SaaS is now my weekend project

How LLM-assisted coding is changing the build-vs-buy calculus for engineers

A surprising number of projects or SaaS can be built in a weekend with LLM assistance nowadays.

Andrej Karpathy called this “vibe coding” when he coined the term in February 2025: talking to an LLM, accepting changes, iterating until the thing works. “Not too bad for throwaway weekend projects,” he wrote. The concept resonated so widely that Collins English Dictionary named it Word of the Year for 2025.

If building software got this much cheaper and faster, what happens to all the subscriptions for software that’s just not that hard to build?

This article is primarily about engineers’ relationship with their software subscriptions. For most users who don’t code, the landscape hasn’t changed much. But for those who can wield an LLM effectively, the economics … is different.

What’s suddenly vulnerable?

A surprising number of paid apps fall into a category I’ll call “CRUD-plus-UX”: they store your data, let you view and edit it, and wrap it in a reasonably nice interface. Think todo apps, markdown editors, bookmark managers, habit trackers, screenshot utilities, personal finance trackers.

These apps typically charge $5–20 per month. They work fine. But for a developer willing to spend a Saturday, they’re replicable.

Productiv’s 2024 report found that the average enterprise used 371 SaaS applications in 2023 but only 220 a year later, a 40% reduction. The stated reason was consolidation and cost savings. Some of that is IT-driven rationalization. Some of it is teams realizing they don’t need specialized tools for everything.

When the alternative to a $15/month app is “i can describe what I want to an LLM and have working code in a few hours and push it on github”, the math is obvious for people who can do that.

Where is the moat?

Yet, this approach can’t replace everything. The real value of a good SaaS product is not just the code. It’s all the data and all the facets around:

• Network effects and integrations. Slack isn’t valuable because of its chat code. It’s valuable because it connects to 2,400 apps and your whole company is already on it. A vibe-coded alternative starts at zero integrations and zero users.
• Multi-user, multi-platform complexity. The moment you need real-time collaboration, RBAC, or native apps across iOS, Android, macOS, and Windows, you’re no longer in weekend project territory. You’re in “maintaining a product” territory.
• Support and accountability. When your weekend project breaks at 2am before a deadline, you fix it. When a paid tool breaks, you open a ticket and someone else’s job is to fix it. That’s worth money when the stakes are real
• Compliance and critical data. If you’re handling health records, financial data, or payroll, you need SOC 2, HIPAA, or PCI compliance. You need audit logs, encryption at rest, backup verification, and someone to call when things go wrong. Your weekend project won’t have any of that.
• Trust and brand. Would you store your passwords in a vibe-coded app? Your medical records? Your financial data? Established vendors have years of reputation, security audits, and bug bounty programs. Your weekend project has none of that.
• Onboarding and documentation. You know how your tool works. Your team doesn’t. Commercial products invest in tutorials, help centers, and UX polish that makes adoption possible for non-builders.

When to build, when to buy, when to deploy

Given these tradeoffs, here’s a practical framework.

Build it yourself when:

• You’re the only user (or your household)
• The data isn’t sensitive or critical
• The problem is narrow and well-defined
• You’ll use stable, boring technologies (files, SQLite, basic web APIs)
• The coding itself brings you satisfaction or learning

Self-host open source when:

• A mature project exists that does 80%+ of what you need
• You’re comfortable with basic server administration
• You want ownership without starting from zero
• Community maintenance has value

Pay for a subscription when:

• Multiple people need access with different permissions
• You need to meet compliance requirements
• The data is irreplaceable or sensitive
• The vendor handles platform churn you don’t want to track
• Uptime matters for business continuity
• Support and documentation have value to you

The sweet spot for vibe coding: glue code and personal utilities. A script that fetches your bank transactions and formats them. A local app that tracks your workouts exactly how you want. A tool that scrapes a forum and sends digest emails. Narrow problems, single user, low stakes.

Where it gets risky: anything involving credentials, PII, or workflows where failure has consequences. Build these if you enjoy the challenge, but don’t do it to save $10/month.

What this means for SaaS builders

If you’re building a product that could be described as “CRUD with a nice UI”, your addressable market among technical users just got smaller.

Engineers who could afford your subscription can now afford to replicate your core value proposition. They were never your most profitable customers, but they were often your most vocal advocates. That channel is narrowing.

What survives this shift:

• Hybrid license models. JetBrains has offered perpetual licenses with annual update subscriptions for years. “Working Copy” on iOS does the same. You pay once for the current version forever, and optionally pay again for future updates. This model respects the “I want to own my tools” instinct while still funding ongoing development. Usage-based pricing is also growing, with firms using pay-as-you-go models reportedly growing faster than subscription-only peers.
• Ops, not code. The value proposition shifts from “we wrote the software” to “we run the software”. Managed databases, hosted CI/CD, monitoring infrastructure: these compete on reliability and operational expertise, not features. An engineer can spin up Postgres. Not everyone wants to be on call for it.
• Content, community, and trust. A note-taking app is replicable. A tax software company’s 30 years of accumulated edge case knowledge isn’t. A decade of earned reputation isn’t. The software becomes the vessel for value that can’t be vibe-coded.

The FTC’s Click-to-Cancel rule is a signal of where regulation is heading: making it easier for customers to leave forces vendors to prove ongoing value.

Owning your stack

The question used to be “can I build this myself?” Now it’s “should I own this?”

For a growing number of engineers, the answer is yes for personal tools and no for critical infrastructure. Your todo app can be yours. Your password manager probably shouldn’t be.

The mediocre middle of software (apps that were fine but not great, that charged $8/month for something you used twice a week, that offered features you didn’t need while missing the one you wanted) is facing pressure from both directions. Above, consolidated enterprise platforms absorb functionality. Below, engineers with LLMs build exactly what they need.

What remains is software that earns its subscription through operational excellence, irreplaceable expertise, or trust built over years. That’s a higher bar than “we built an app.”

Originally published on my substack: https://sderosiaux.substack.com/p/your-saas-is-now-my-weekend-project