04 May Stop Building Software. Start Replacing Work.
The Next €100B Companies Won’t Sell Software. They’ll Deliver Outcomes.
The signal is no longer just noise; it’s a roar. By early 2026, industry estimates suggested companies like Anthropic had moved from roughly ~$1B revenue run rate at the start of 2025 to a $30B revenue run rate end of March 2026. While the numbers are staggering (there’s no equivalent in economic history), the ambition behind them is more revealing. When leaders such as Dario Amodei publicly target $100B in ARR towards the end of this year (!) and even $1T in ARR over the next 3 years, they aren’t describing a typical software company.
They are describing a system designed to absorb parts of the $70T global service economy.
This shift is happening now for a simple reason: these systems deliver orders-of-magnitude gains in productivity (an explosion in economic value creation). For the first time, software can both reason and act. Large models, combined with agentic workflows, allow systems not just to generate answers, but to execute tasks across tools, data, and processes.
The Shift: From Tool-Maker to Work-Taker
Historically, the value proposition of software was simple: give a human a better tool so they can work faster. CRMs made sales reps more organized; ERPs made resource planning more legible. This was the era of digitizing workflows.
We have now entered the era of displacing workflows. AI systems will no longer be selling “assistance”; they will be selling “outcomes.”
- Old World: A tool that helps a lawyer review a contract.
- New World: A system that returns a fully analyzed, risk-mitigated contract.
This distinction explains the magnitude of the projected revenue. SaaS competed for software budgets (a few percentage points of revenue). AI-native companies are competing for payroll and outsourcing budgets—the largest line items on any P&L.
The Structural Productivity Shock
This transition is triggering a series of structural shocks across labor-heavy industries. In customer support, agents aren’t just managing tickets; they are resolving them instantly. In software development, build cycles are collapsing from weeks to hours.
We are moving away from incremental improvements toward a “Value per Customer” metric that is orders of magnitude higher than anything seen in the SaaS era. When a startup can eliminate millions in labor costs for a client, their pricing power ceases to be tied to “per-seat” licenses and begins to reflect a share of the total economic value created.
The “Harness” as the New Moat
A common skepticism persists: “If everyone has access to the same foundation models, where is the defensibility?”
This misses the point of where value actually accrues. Access to a model is not a business; execution is. The durable winners won’t be defined by the underlying LLM they use, but by the “Harness Layer” they build around it.
The Harness Layer is where raw model capability is converted into deterministic business outcomes.
Think of it this way: everyone now has access to the same underlying intelligence. But access is not advantage. Just as using the same racket as Novak Djokovic doesn’t make you play like him, access to powerful AI models does not create value on its own.
The difference is not the tool. It is how the tool is operationalized.
The Harness Layer is the “how”: the system that turns raw AI capability into reliable, repeatable execution inside real-world environments. More on this harness concept here: https://x.com/rohit4verse/status/2033945654377283643?s=46
It is also where switching costs, data accumulation, and workflow lock-in emerge.
The difficulty of building this layer is exactly where the moat emerges. It is the “glue” that turns raw intelligence into a reliable service.
A New Framework for Founders and Investors
For those building or backing the next generation of giants, the evaluation framework must shift. The relevant question is no longer “What software are you building?” but “What service are you eliminating?” and “What outcome are you delivering?”.
For Founders: Stop building features. If a human still has to execute the core work, you are under-delivering. Design for “automation-first,” where the human moves from the operator to the supervisor. Models may not allow you to do this today but they will in months. Be ready to deliver the work and the outcome your customers want.
For Investors: Look past the UI and CAC. The new benchmark is:
- What % of a workflow is automated?
- How many euros of labor are displaced per customer?
- How many euros of labor are eliminated per euro of revenue?
Can this company scale its impact without a proportional increase in its own (or its customers’) headcount?
The Path to Trillions
We aren’t just entering another software cycle; we are witnessing the birth of AI-native replacements for high-cost services. While revenue ceilings expand dramatically, margins will concentrate in companies that own the harness layer and the workflow—not those reselling commoditized model output.
This creates a massive opening for focused ecosystems. When intelligence becomes widely accessible, the advantage shifts to those with the deepest understanding of complex, real-world problems and the speed to build the harness around them.
The companies that move from millions to billions—and eventually trillions—won’t be the ones that sold the best tools.
They will be the ones that absorbed complexity, automated it, and sold the outcome.