SaaS is Dead. Here's What Comes Next.

By SpecMarket Team
SaaS is Dead. Here's What Comes Next.

$1 trillion in SaaS market cap evaporated in a single week. Forrester called it “the great unwinding.” We call it overdue.

For decades, the software industry has operated on a simple premise: rent, don’t own. Pay $25/seat/month for e-signatures. $75/seat/month for CRM. $10/user/month for a glorified to-do list. Multiply by every employee, every month, forever.

The average mid-size company spends $4,830 per employee per year on SaaS subscriptions. That’s Forrester’s number. For a 50-person team, that’s $241,500 annually — on software you don’t own, can’t modify, and lose access to the moment you stop paying.

That model is dying.

What changed

Two things happened in the last 18 months that broke the economics of SaaS rentals.

First, AI coding agents got good enough to build real software autonomously. Not demos. Not scaffolds. Working applications with tests, error handling, and deployment configurations. Claude Code, Cursor, Windsurf — the tooling crossed a threshold where a well-specified task reliably produces working software.

Second, the cost of running these agents dropped. A year ago, building a non-trivial application with an AI agent cost hundreds of dollars in API tokens and took days of babysitting. Today, a well-written specification fed to Claude Code produces a working e-signature application in 34 minutes for $11.40 in tokens. A form builder costs around $9. A basic CRM, $25.

Compare those numbers to annual SaaS costs for a 20-person team:

  • E-signatures: $6,000/year vs $11.40 once
  • CRM basics: $18,000/year vs $25 once
  • Form builder: $6,000/year vs $9 once
  • Scheduling tool: $2,400/year vs $6 once

“Once” is the operative word. You run the spec, you get the code, you own it. No recurring charges. No per-seat pricing. No pricing page that mysteriously changes every quarter.

The Ralph Loop changes everything

This is the mechanism. A Ralph Loop is an autonomous agentic coding loop: you feed a detailed specification to an AI coding agent, and the agent reads it, writes code, runs tests, hits errors, fixes them, and iterates — without human intervention — until the software works or it gets stuck.

The name comes from Geoff Huntley’s work on autonomous coding workflows. The insight was deceptively simple: the bottleneck in AI-assisted coding isn’t the AI’s capability. It’s the quality of the instructions. A well-specified task produces reliable results. A poorly-specified task produces garbage.

Ralph Loops formalise this insight into a repeatable process. A spec is a structured directory with a functional specification, technology stack requirements, code patterns, security rules, and machine-evaluable success criteria. The agent reads the spec, builds the software, and checks its work against the criteria. No ambiguity. No “it works on my machine.” Either the criteria pass or they don’t.

When one person runs a spec successfully, that’s a data point. When 230 people run the same spec and 94% succeed across different machines, operating systems, and network conditions — that’s a validated blueprint. The failures teach the spec author where the gaps are. The successes give new users confidence that the spec delivers what it claims.

What SpecMarket is

SpecMarket is the marketplace for these specifications. Think of it as the npm for SaaS replacements.

Engineers publish specs — battle-tested blueprints for replacing specific SaaS tools. The community runs them. Every run is tracked (with user consent): success rate, cost, build time, which success criteria passed and which failed. All of this data is public. A spec’s reputation is earned through transparent, verifiable results — not marketing copy.

There’s also a bounty system. Companies or individuals post bounties for specs that replace specific SaaS tools (“$500 for a spec that replaces Calendly with 90%+ success rate”). Creators build specs that meet the bounty criteria, submit their work, and earn money when the community validates it.

The whole thing runs on three principles:

Transparency. Every spec’s source is readable before you run it. Every run’s metrics are public. If a spec only handles 60% of what Notion does, the marketplace says that — not “full Notion replacement.” We’d rather lose a user today and keep their trust than oversell and lose it forever.

Ownership. The code a spec produces is yours. Deploy it on Vercel, Railway, your own server, or a Raspberry Pi in your closet. Modify it. Extend it. Hand it to a contractor. You paid for the API tokens; the result belongs to you.

Honesty about limitations. Not every SaaS tool can be replaced by a spec. Products with deep network effects (Slack, LinkedIn), heavy regulatory compliance requirements (certain fintech tools), or real-time collaboration at scale (Figma) have genuine moats. We say so explicitly. The “SaaS is dead” thesis applies to commodity workflows — the 80% of tools where you’re paying $25/seat/month for something an engineer could build in an afternoon with the right instructions.

The nuance matters

SaaS isn’t dead for everything. That headline gets attention, but the reality is more precise.

SaaS is dead for commodity workflows that can be spec’d. Document generation. Form handling. Basic CRM. Scheduling. Internal dashboards. Invoice generation. These are well-understood problems with well-defined inputs and outputs. They don’t benefit from network effects. They don’t require real-time collaboration between thousands of users. They’re perfect targets for specification-driven builds.

SaaS is still alive for products with genuine competitive moats: deep integrations with regulated systems, massive network effects, proprietary data, or real-time collaboration that requires dedicated infrastructure. We’re not here to pretend otherwise.

But here’s the thing — that “commodity” category is enormous. It represents hundreds of billions in annual SaaS spending. Most of it is pure rent-seeking on solved problems. The specifications for these tools aren’t complicated. They’re just not written down in a format that AI agents can execute reliably. Until now.

What happens next

We’re building SpecMarket in public, using the same Ralph Loop process we’re building the marketplace to support. Our specs are open. Our costs are tracked. Our metrics will be transparent.

If the process works, you’ll see it in the numbers. If it doesn’t, you’ll see that too.

The marketplace is pre-launch. The CLI is being built. The first specs are being written. If you’re paying $500/month for SaaS tools that handle commodity workflows — document generation, form handling, scheduling, basic CRM — the specifications to replace them are either already written or being written right now.

Install the CLI. Browse the marketplace. Pick one spec. Run it.

The worst case: you spend $10 on API tokens and learn something about how autonomous coding works. The best case: you stop renting software you could own.

SpecMarket is being built in public. Follow along on Twitter/X or install the CLI to try it yourself.