SaaS unbundles

There's a thing happening to SaaS that mostly gets read as "AI will eat it" or "no it won't". Both are wrong. SaaS is unbundling.

The pitch always had two layers stacked together. The bottom was a backbone — identity, audit logs, role-based permissions, integrations with the other forty systems, compliance certifications, support contracts, uptime SLAs, change management. The top was the value — the workflow you'd otherwise build yourself. Same vendor, same contract, same bill. You couldn't realistically separate them, so the bundle held.

Two things changed.

Vibe coding made small workflow tools genuinely buildable by anyone with judgement and a working AI. Not toys — real tools, used in production, by people who know what good output looks like. Done by an expert it works. Done without that judgement it becomes somebody's pet, the way unmaintained spreadsheets always did.

Agents made the workflow layer composable in a different way. You assemble the behaviour you want from APIs, prompts, and policies, rather than picking a vendor whose screens happen to fit.

Both moves do the same thing to the bundle. They make the top layer — the actual value the customer extracts — cheap to rebuild, configure, or replace. What stays expensive is the backbone. Nobody vibe-codes SOC 2. Nobody composes their way to forty integrations and a 99.95% SLA in a weekend. The backbone is the part of SaaS that has always been hard, and it remains hard.

So the bundle splits. Backbone migrates toward infrastructure — priced like Stripe or Twilio, sold on reliability and reach, defended by network effects on the integration side. Value layer becomes a composition problem — sometimes vibe-coded by a domain expert, sometimes assembled from agents, sometimes still bought from a vendor when buying is cheaper than composing.

Pricing splits the same way. Backbone commands rent because the option to rebuild it doesn't exist. Value-layer SaaS — the workflow tool with a thin compliance moat — runs into a customer base that can plausibly rebuild the 20% of features they actually use. That's a different negotiation than it was five years ago, even for customers who never actually exit.

This applies most cleanly to enterprise workflow SaaS in the $20–200 per seat band. Not everywhere. SMB tools, true platforms, and SaaS that owns a measurable outcome behave differently. But the squeezed quadrant — pure value layer, no backbone, no outcome accountability — is real, and getting more real.

The catch on the buyer side is that composing the value layer well requires knowing what good looks like. Pick the wrong workflows to automate, give the agent the wrong policy, and the composition rots in a way that's hard to see — the surrounding setup looks structured, the outputs keep flowing, but the quality drifts. The barbell middle, again. Composability without domain expertise is just faster mistakes.

The practical read is simple. For buyers: pay for the backbone, compose the value layer with people who actually know the domain. For SaaS founders: own a backbone or own an outcome. Owning neither is the squeezed seat, and vibe coding plus agents are not going to be kinder to it next year.

If you've got SaaS line items where the entire value is a workflow, AI can probably replace them now. Reach out — happy to work out which ones are worth replacing, and how to do it without breaking the backbone you actually need.

Update: Backbone is contested, not safe sharpens this — outcome ownership is the durable position; backbone is a category fight with three winners per slot.