Does Arco avoid all SaaS products?

No. The distinction is between Application SaaS and Infrastructure SaaS. Infrastructure SaaS — cloud compute, raw LLM APIs, database services, network primitives — is priced by consumption, optimised for machine throughput, and has no UI overhead built into its cost. Arco uses it as a utility. Application SaaS — tools designed for human collaboration, task management, and data entry, priced per named user and optimised for screens — has no place in an agentic stack. The question Arco applies to every tool is: does this thing's value live in its interface? If yes, it is Application SaaS and the answer is no. The Stewards who govern Arco's agentic stacks may use lightweight monitoring or governance tools where genuinely useful — but these are evaluated by the same criterion: priced by consumption or flat-rate utility, never by named user seat.

The UI Tax is the cost premium embedded in SaaS products designed for human use. It includes the graphical interface development overhead, the permissions and access management infrastructure, the user training resources, and the subscription premium charged for enterprise UX. Agentic workflows require zero UX for routine execution. An agent calls an API and receives a structured response. The graphical layer, the access controls, the onboarding flow — these are built for human operators and priced accordingly. De-SaaS-ing eliminates the UI Tax by connecting directly to the data and logic layer rather than the human-facing application built on top of it.

Do Stewards use software?

Yes — but the footprint is radically smaller than a traditional manager's, and the selection criteria are different. A Steward monitoring Execution Divergence, reviewing Ghost Trial outputs, or processing an exception queue may use lightweight tooling to do so. What a Steward does not use is Application SaaS priced per seat for tasks the agentic system now runs autonomously. The Steward is not doing tasks — they are governing a system. The software that supports governance is evaluated by the same standard as everything else in the stack: priced by consumption, not by headcount. A traditional manager's software spend scales with the number of people they coordinate. A Steward's software spend scales with the complexity of the system they oversee, which for a well-architected autonomous business is low and falling.

How does De-SaaS-ing affect the speed and cost of a build?

In the short term, it requires more engineering investment. Building a direct database integration takes longer than subscribing to a SaaS tool with a preconfigured connector. In the long term, it is faster, cheaper, and structurally superior. Arco is not locked to any vendor's roadmap, pricing decisions, or API deprecation schedule. The Architectural Certainty of the system is maintained because the logic layer is owned, not rented. And through the Arco Flywheel , every direct integration built for one company reduces the engineering cost of the equivalent integration in the next. The first build absorbs the overhead. Subsequent builds inherit the solution.

How does Sovereign Infrastructure relate to the Arco Flywheel?

Directly. Every De-SaaS-ing decision produces an owned integration or a validated open-source configuration that becomes part of the Agentic Core shared across all Arco companies. A direct API integration built for a logistics deployment is the template for the compliance integration. A database connection pattern validated in one market is the starting architecture for the next. Sovereign Infrastructure does not just reduce costs on a per-company basis — it reduces the engineering overhead of every subsequent build by eliminating repeated vendor evaluation, negotiation, and integration work. The infrastructure compounds in the same way the logic does.

Does this model affect how an Arco business is acquired?

Significantly. An acquirer taking ownership of an Arco business does not inherit a portfolio of SaaS vendor relationships, seat licence renewal cycles, or subscription dependencies managed by a procurement function. They inherit documented integrations, owned logic, and compute costs that are legible and optimisable from the first day of ownership. This is a direct contributor to the 70% PMI timeline reduction Arco targets. A SaaS-dependent business requires the acquirer to audit, renegotiate, and either retain or replace every vendor relationship in the stack. A Sovereign Infrastructure business requires a technical review of owned integrations. The difference in due diligence burden is substantial.

What is the relationship between De-SaaS-ing and the Machine-Readable Business?

They are complementary architectural principles. De-SaaS-ing governs how Arco's operating companies consume software — inbound, as buyers of infrastructure. The Machine-Readable Interface governs how they present themselves to the market — outbound, as providers discoverable by agents. Both require an API-first architecture. A business that has De-SaaS-ed its internal stack has already built the discipline required to construct a proper MRI layer: it understands how to interact with data through structured interfaces rather than human-mediated applications. The two principles reinforce each other at the infrastructure level.

Why Autonomous Businesses Don't Buy SaaS

The SaaS model is a tax on headcount. For an autonomous business, it is an operational liability. Traditional software-as-a-service is priced on the assumption of human labour: per-user seat licences are designed for organisations that scale by adding people. If your business model requires a hundred people to manage a process, you pay for a hundred seats. If you deploy agents to run that process instead, and continue paying for the same seats, you have not captured the value of the architecture. You have transferred your payroll to a software vendor.

Arco does not pay seat licences for processes agents now run. We build infrastructure for agents and equip the Stewards who govern them with tools priced by consumption, not by headcount.

Why the SaaS Model Fails for Agentic Operations

The SaaS industry built its pricing model on a reasonable assumption: that the primary user of software is a human who needs a graphical interface, permission controls, a dashboard, and a support tier. Every element of the per-seat cost structure exists to serve that human. The interface development. The access management layer. The user training programmes. The enterprise support contracts priced per named user.

An agent needs none of this. An agent needs an API endpoint, a structured data contract, and a response within a defined latency threshold. It does not log in. It does not navigate a dashboard. It does not require onboarding. The entire UI Tax — the cost of software designed for human cognition rather than machine execution, including the GUI development overhead, the permissions infrastructure, and the UX premium baked into subscription pricing — is dead weight in an agentic stack. Arco does not pay it. Most firms still do, even after they have deployed the agents that make it redundant.

According to McKinsey's State of AI (2025), 88% of organisations have adopted AI in at least one business function, but only 6% qualify as high performers generating measurable EBIT impact. The defining characteristic of that cohort is workflow redesign — fundamental reconstruction of how the process runs, not a tool layered on top of the existing structure. Paying for SaaS seat licences after deploying agents is the opposite of workflow redesign. It is paying for the structure you were supposed to replace.

Legacy firms pay for seats. Arco pays for compute.

De-SaaS-ing: The Architecture of Compute-Based Costs

De-SaaS-ing is Arco's operational discipline of replacing per-user subscription software with API-first, compute-based infrastructure — direct integrations with the underlying data and logic layers rather than the human-facing application built on top of them. Where a traditional firm subscribes to a CRM to manage customer data, an Arco business connects directly to the database. Where a traditional firm pays per seat for a workflow tool, an Arco business builds the workflow as an agentic loop with no UI layer required for routine execution.

The measurable consequence is a 65% reduction in internal software spend compared to human-centric competitors at equivalent revenue. That figure reflects the elimination of the UI Tax across the tooling stack: the interface-heavy SaaS products that charge $50 to $200 per user per month for capabilities an agent accesses in a single API call. Each eliminated subscription compounds into the Operational Drag reduction that makes the autonomous business's cost structure structurally superior to the incumbent's, not temporarily cheaper.

The distinction Arco draws is between Application SaaS and Infrastructure SaaS. Application SaaS — tools designed for human collaboration, data entry, and task management — is priced for seats and optimised for screens. Arco does not use it. Infrastructure SaaS — cloud compute, raw LLM APIs, database services, network primitives — is priced by consumption and optimised for machine throughput. Arco uses it exclusively and treats it as a utility to be queried and replaced when a more efficient option emerges. The rule is simple: if the tool's value is primarily in its interface, it has no place in an agentic stack.

Sovereign Infrastructure

Sovereign Infrastructure is Arco's architectural standard for the software layer underneath its operating companies: wherever possible, the agentic core runs on open-source models and direct-access databases rather than third-party application platforms. Arco owns the logic. External infrastructure is rented by consumption. No vendor roadmap determines what an Arco company can or cannot do next quarter. No seat-based pricing model grows in proportion to operational success.

The cost structure this produces is qualitatively different from a SaaS-dependent competitor. A SaaS-heavy firm has a fixed floor of subscription costs that does not decrease when revenue contracts and does not scale proportionally when revenue grows — it scales with headcount, which for a legacy firm is a proxy for revenue but for an autonomous business is not. Arco's costs scale with compute consumption, which scales with actual business activity. The cost of goods sold is tied to what the business produces, not to how many named users are logged into the platform.

This also directly enhances exit value. An acquirer taking ownership of an Arco business does not inherit a stack of vendor relationships, seat licence agreements, and renewal cycles managed by a procurement team. They inherit documented integrations, owned logic, and compute costs that are legible and optimisable from day one. Sovereign Infrastructure is a component of Turnkey Margin: the business is transferable as a technical operation because the software layer was never held together by human-managed subscriptions.

Every De-SaaS-ing decision Arco makes also compounds through the Arco Flywheel. The direct API integration built for a logistics deployment becomes the template for the compliance build. The open-source model configuration validated in one market is the starting point for the next. Sovereign Infrastructure does not just reduce costs on a per-company basis — it reduces the engineering overhead of every subsequent build by eliminating the need to re-evaluate, re-negotiate, and re-integrate the same vendor relationships from scratch.

The Operator's Verdict

The Coordination Tax that makes incumbents structurally slow manifests at the software layer too. Every SaaS seat licence is a payment for the coordination infrastructure the incumbent cannot remove: the approval workflows, the dashboard reviews, the human access management. An autonomous business that still pays for these is not autonomous — it is automated with a smaller team sitting inside the same cost structure. The transition from headcount to compute is not complete until the software spend reflects it. If you are still paying for seat licences in an agentic workflow, you are not running an autonomous business. You are subsidising a legacy software vendor. The most profitable companies of the next decade will be those that own their logic and rent only their compute.

Related Operational Memos

Memo #03: Overhead Is a Design Choice — How stripping legacy software overhead is a core architectural requirement, not a cost-cutting measure.

Memo #13: The Machine-Readable Business — Why the same API-first discipline that enables De-SaaS-ing also makes Arco's businesses discoverable to agents.

Memo #06: Legacy Liability — Why incumbents cannot De-SaaS their stacks without dismantling the human-centric architecture they were built on.

KEY TAKEAWAY

How do autonomous businesses manage their software costs?

Autonomous businesses eliminate per-user SaaS seat licences by replacing Application SaaS with API-first, compute-based infrastructure. This is Arco's De-SaaS-ing discipline: direct integrations with underlying data and logic layers rather than the human-facing application built on top. The UI Tax — the cost premium baked into SaaS products for graphical interfaces, permissions management, and human UX — is redundant in an agentic stack. Agents need API endpoints, not dashboards. The measurable result is a 65% reduction in internal software spend compared to human-centric competitors at equivalent revenue. Arco's cost of goods sold scales with compute consumption, not headcount. Key metric: 65% reduction in internal software spend vs. human-centric competitors at equivalent revenue.

The Death of the Seat License: Why Autonomous Businesses Don't Buy SaaS

Why the SaaS Model Fails for Agentic Operations

De-SaaS-ing: The Architecture of Compute-Based Costs

Sovereign Infrastructure

The Operator's Verdict

Related Operational Memos