The IaaS/PaaS/SaaS Singularity

Ask why the cloud has three layers and you'll get a technology answer: IaaS is VMs, PaaS is the runtime, SaaS is the application. Stack them like a diagram, each one abstracting the one below.

The diagram is wrong, or at least it names the wrong thing. What actually separates the layers isn't technology. It's labor — specifically, whose labor operates each slice of the stack:

• IaaS: you rent resources; you supply all the operating expertise. Pricing follows: pure metering, per resource, per service. Their labor cost is near zero, so the price is the resource.
• PaaS: they operate the middle — the runtime, the deploys, the scaling — and you pay a convenience markup on every resource that passes through. The markup is the wage of the platform team you didn't hire.
• SaaS: they operate the entire application, and you pay per seat — rent on software you can't run yourself, priced not by what it costs to serve you but by what your alternative (building and operating it) would cost.

Three layers, three pricing models, one underlying variable: how much of someone else's operational labor you're buying. The cloud's entire commercial structure is a labor market wearing a technology costume.

Which means the interesting question isn't "what happens when the technology gets better?" It's: what happens when the labor collapses?

The labor is collapsing

That's not a forecast; it's what k8gentic engineering is. An agent that can operate canonical Kubernetes — read the manifests, drive the reconcilers, verify the rollout — is operating expertise, on tap, at machine speed. The substrate grew trusted loops at every stage precisely so that an agent could run the real thing instead of a hosted imitation of it.

Walk the layers with that in hand:

The PaaS layer dissolves first. Its entire value was operating the middle so you didn't have to. But the middle is now canonical primitives — Helm, Flux, Gateway API, CRDs — and the agent operates them directly. The convenience markup has nothing left to mark up. This is the economic reading of the PaaSocalypse: not that PaaS gets cheaper, but that the labor it was reselling stops being scarce.

Then the SaaS layer thins. Per-seat rent was priced against your alternative: building and operating the application yourself, which used to be absurd. But agents operating open-source applications — a CRM, an analytics stack, a support desk, deployed and run on your own substrate — make the alternative cheap. Not free: the software still has to run somewhere, still needs storage and memory and egress. But the rent component — the part of the seat price that was scarcity, not service — deflates toward zero. What survives of SaaS is the genuinely hard hosted service; what dies is seat-rent on commodity workflows.

And IaaS? IaaS was always the honest layer — meter the resources, supply no labor. The singularity doesn't kill it; it promotes it. When the labor at every layer above is an agent, resources are the only real cost left in the whole stack.

Three layers, collapsing into one: resources, operated by agents, composed from open source. That's the singularity. Not a merger of product categories — an evaporation of the labor differences that made them separate categories in the first place.

What's left to price

Here's where this stops being commentary and becomes a pricing model, because we had to answer this question with a number on a page.

If the layers were labor, and the labor is now an agent, then almost everything that used to justify a margin is gone. You can't charge a convenience markup for operating the middle — the customer's agent operates the middle. You can't charge seat-rent for the application — the customer can deploy the open-source equivalent in an afternoon. What's actually left?

1. The resources. Real, measurable, unavoidable. CPU, memory, storage, egress.
2. A thin service margin for the things that are still genuinely ours to do: keeping the substrate healthy, the catalog curated, the provisioning instant, the platform boringly reliable.

So that's the price: measured cost + 35%, across the board, published. Not because we're generous — because in the singularity, that's all the margin that's defensible. Every classical cloud margin was rent on a labor asymmetry, and the asymmetry is gone. Our customers run the same open-source kilter stack we run. They can self-host it. They can move to another Miniscaler running it. When your customer can replicate your entire value chain, the only pricing that survives contact is the honest kind. Sovereignty isn't a feature we offer despite our margins — it's the mechanism that disciplines them.

And per-seat pricing? It survives exactly where per-seat value genuinely lives: humans. The Kilter Business Portal — where non-engineering people approve agent actions, operate workflows, see what's running — is priced per user, because its value really does scale with the humans using it. Infrastructure isn't. An engineer's fifteen apps and four hundred agent runs cost us compute, not seats — so seats are the wrong meter, and we don't use them. One pricing principle, applied twice: meter what actually costs, charge for where value actually accrues, and never confuse the two.

The full model is on the pricing page — flat platform fee, included allocations, cost+35% above them, portal seats from $5. You'll notice it reads less like a cloud rate card and more like a bill of materials. That's the singularity showing through.

The stack after the singularity

The IaaS/PaaS/SaaS taxonomy had a good twenty-year run, and it earned it — labor really was that scarce, at every layer, for that long. But price is where taxonomies go to die. When an agent operates canonical Kubernetes, the PaaS markup has no labor to resell. When an agent operates open-source applications, the SaaS seat has no scarcity to rent. What remains is the substrate, the resources, and a thin honest margin for keeping the lights reliably on.

Three letters-as-a-service, collapsing into one stack you can actually own. The companies still priced for the old taxonomy are betting the labor stays scarce. We're betting it already isn't.

Where to go next

1. Read the PaaSocalypse — the fragmentation bill that pushed the middle layer to the brink.
2. See The k8gentic Revolution — Kubernetes made simple, not easy: the agent-capability bet that collapses the labor.
3. Read Building Agents on Open-Source Apps — how the SaaS layer thins in practice.
4. Then look at the pricing this argument produces — and start the trial if it holds up.