AI is not a tool.
It is the substrate of the next operating layer.

For the past two years, every company on earth has been adding AI to its operations. Copilots in the IDE. Assistants in the CRM. Summarizers in the inbox. The gains have been real — measurable productivity, clear ROI, satisfied executives. And yet none of it has changed the shape of how companies actually work. The fragmented software, the manual handoffs, the workflows designed in 2009 — all of it is still there, with AI bolted to the side.
This essay is an argument for what comes next, and what Pragma is building toward.


I. The premise
The next decade's category leaders will not be the companies with the most AI tools. They will be the companies that have rebuilt their operations around AI as a substrate.
A substrate is the layer beneath the layer — the foundation other things sit on. Electricity was a substrate. The internet was a substrate. Cloud compute was a substrate. Each one didn't just improve existing businesses; it forced a redesign of what a business was. The companies that won in each of those transitions weren't the ones who adopted the new technology earliest. They were the ones who restructured around it most completely.
Amazon didn't become Amazon by putting a website on top of a mail-order catalog. It became Amazon by rebuilding the entire logistics, inventory, and customer experience around the assumption that the internet existed. The companies that built websites on top of catalog operations are mostly forgotten now. Different layer of restructure, different outcome.
AI is the next substrate. And almost no one has restructured.


II. What's actually broken
The fragmented-software era — roughly 2010 to 2024 — was a coping mechanism for a deeper problem: companies couldn't afford to build coherent operational systems, so they bought twenty disconnected ones. Salesforce for sales, Zendesk for support, NetSuite for finance, Asana for tasks, fifteen others for everything else. Each tool was a partial fix, glued together by humans doing the work the software couldn't.
That stack made sense in a world where the marginal cost of human judgment was the binding constraint. It does not make sense now. The cost curve has bent. Judgment, execution, and orchestration are becoming machine-priced. The fragmented stack — and the human labor holding it together — is now the inefficiency, not the solution.
Most companies have not noticed yet. They are adding AI to each tool in the stack, one tool at a time, expecting the sum to be coherent. It will not be. You cannot reach AI-native operations by adding AI features to non-AI-native software. The substrate is wrong. The architecture is wrong. The whole stack was built for human-only execution and it shows.


III. The redesign
Operations rebuilt around AI as a substrate look different from operations with AI bolted on. Three things change.
Workflows collapse. Twelve-step processes designed to route work between humans become two-step processes designed to route work between systems. The handoffs that used to take hours take seconds. The forms that used to need filling don't exist. The reports that used to require synthesis are generated continuously. The work itself shrinks.
Roles change. The work that used to require five people doing routine execution requires one person directing the system that does the execution. The remaining humans move up the stack — into judgment, design, and exception handling. The org chart compresses. Headcount stops being a proxy for output. The new question isn't how many people do we need? but how many operators do we need, and who's running them?


Economics change. Cost per task collapses. Throughput compounds. The unit economics of operations-heavy businesses — services, support, finance, ops — start to resemble software. The companies that achieve this shift first will set new price floors that the others cannot match. The competitive landscape stops being about who has the best people and starts being about who has the best operating layer.
This is not a tooling upgrade. It is a different kind of company.


IV. Why it isn't happening
If the redesign is so obvious, why is almost no one doing it?
Because operational redesign is hard, technical, and risky in a way that buying tools is not. It requires three things most companies don't have together: a deep understanding of the existing operation, the engineering capacity to rebuild it, and the operational discipline to run the new system in production.


Consulting firms have the first. They can map your operation in a forty-slide deck and tell you where the bottlenecks are. But they don't ship systems. They ship recommendations.


Software vendors have the second. They can build powerful tools and sell you seats. But they don't run your operation. They sell you software and let you figure out the deployment, the change management, the integration, and the ongoing optimization.
AI agencies and automation shops are emerging to fill the gap, but most of them are either too small to handle real operational scope or too focused on individual workflows to redesign the whole stack.


Nobody has all three things together at scale: the analytical depth of consulting, the engineering capacity of software, and the operational accountability of an embedded team. The result is a market full of advice that doesn't ship and tools that don't integrate. Companies are stuck buying point solutions because the alternative — actually redesigning operations — has no credible provider.


V. What Pragma is
Pragma is what we believe a credible provider looks like.
We are an operational infrastructure company. We work with companies that have outgrown the tooling their operations were built on, and we redesign how they execute work.


We map the existing operation — every workflow, every handoff, every cost center. We specify the AI operators, intelligent workflows, and infrastructure required to rebuild it. We deploy the systems into production. And our embedded execution teams run them alongside yours until the new operation is faster, leaner, and more accountable than the old one.


We do not sell software. We do not sell advice. We sell operational outcomes — and we own them.
Concretely, that means three things you can buy from us today, at three different depths.


You can deploy a single Operator — one AI operator running one workflow in your business, live in two to four weeks. This is where most companies should start: pick the workflow costing you the most, prove the model works, expand from there.
You can deploy a System — a connected set of operators and workflows redesigning a whole function. Sales, support, finance, or operations, rebuilt around AI as a substrate rather than bolted on top of legacy software.


Or you can deploy the Backbone — Pragma running as the operational backbone of your business. Multiple connected functions, embedded team, accountable for outcomes month over month. This is what an AI-native company looks like.
All three are real engagements. All three deliver the same thing: a running operation, measured by outcomes, owned by Pragma.


VI. What we believe
Five things, briefly.

  1. AI is a substrate, not a feature. Treating it as a feature produces bounded gains. Treating it as a substrate produces companies that operate differently from their competitors at every level.

  2. The next category is operational infrastructure. Not vertical SaaS. Not horizontal copilots. The layer that sits beneath operations and runs them — the AWS of how businesses execute work.

  3. Services and software are converging. The companies that win will own both — the system and the people running it. Pure-play software vendors will be out-executed by integrated providers. Pure-play consultancies will be out-shipped by operators. The middle ground that didn't exist five years ago is the only ground worth standing on now.

  4. Ownership of outcomes is the only defensible position. Anyone can ship a tool. Anyone can deliver a deck. The companies that take responsibility for whether the operation actually works will compound trust and pricing power that the others cannot. Outcome ownership is the moat.

  5. The window is now. The companies that begin restructuring in the next twenty-four months will set the operational standard for their industries. The companies that wait will spend the rest of the decade catching up. This is not a hypothetical timeline — it's how every substrate shift has played out, and there's no reason to think this one will be different.


VII. The work ahead
Pragma is at the beginning. The companies we work with today are early adopters of an idea that will be obvious in five years and unavoidable in ten. We are building the infrastructure, the operator network, and the orchestration layer that will let any operations-heavy business make this transition — without having to assemble the consulting, engineering, and operational disciplines themselves.


The long-term vision is simple: Pragma becomes the operational layer for modern businesses. The thing companies plug into the way they plug into AWS, or Stripe, or Snowflake — but for how the work actually gets done.


We are not there yet. We are a small team in Helsinki, working with a small number of companies, building what we believe will be the most important operational category of the next decade. We are selective about who we work with, and we expect the same in return.


If you run a company that is feeling the limits of its current operational stack — if you sense that adding more tools will not get you where you need to go — we should talk.


The substrate is shifting. The companies that move first will compound for a decade. We're the team that's going to help you do it.

Redesign your operations. Run them in production.

Redesign your operations. Run them in production.

The operational layer for AI-era businesses.

Helsinki · Available globally

The operational layer for AI-era businesses.

Helsinki · Available globally

The operational layer for AI-era businesses.

Helsinki · Available globally