The Great Decoupling: Building the Atomic Company
How AI agents and service-as-software are weakening the old link between revenue growth and headcount growth, and what it means to build the atomic company.
For decades, executives lived under one rule of scale: revenue growth eventually demanded headcount growth, and a thick layer of coordination to manage it. That rule is weakening fast as AI agents collapse the old boundary between software, meaning a tool, and services, meaning labor. Instead of purchasing a seat, enterprises are increasingly purchasing a capability that executes work. It is arriving with a new operating model: service-as-software.
What changed: capability provisioning replaces tool deployment
We used to buy tools and ask people to do the work.
Now we increasingly buy systems that do the work, with humans managing exceptions, quality, and risk.
Customer operations is one proof point. Vendors like Sierra are explicitly selling agent-delivered experiences, not seat licenses. Operators like Klarna have publicly claimed large volumes of customer-service work absorbed by AI, while still acknowledging edge cases that require skilled humans.
The ladder: from tools to outcomes to elimination
Many enterprises are stuck in two traps.
The first is the seat trap: buy more copilots. Better drills.
The second is the labor trap: outsource more work. More drillers.
Winners move up the ladder:
- Build: in-house drills.
- SaaS: a drill per employee.
- BPO: a driller per process.
- Service-as-software: pay for the hole, meaning an outcome unit.
- Elimination: realize you did not need the hole, then change the system.
Automating the friction is good.
Eliminating the reason friction exists in the first place is better.
The end state: the Atomic Company
The Atomic Company is different from the familiar trope of a small team plus lots of AI.
It is a fully integrated production system across three layers.
1. Core: intent and capital
This is where strategy, product, and investment decisions sit.
The core decides what the company exists to do, where capital should be allocated, and which capabilities deserve to be owned versus rented.
2. Governance layer: the new middle
This is the orchestration layer: QA, security, audit, exception handling, policy, and control.
This is why the middle does not disappear. It mutates.
The work shifts from people management to workflow governance.
3. Agent supply chain: the capability cloud
This is the modular services layer executing workflows.
The organization increasingly provisions capabilities rather than deploying tools or hiring bodies.
The metric: revenue per employee, with guardrails
Revenue per employee is becoming a board-level signal for AI leverage, but it must be read alongside margin, quality, and risk.
If your AI program only adds tools while exceptions, rework, and compliance risk pile up, your KPI will lie.
The goal is not simply fewer people.
The goal is capability without complexity.
Monday morning plan: three atomic questions
1. Where are we buying seats for non-core work?
Audit SaaS by seat-to-outcome mismatch.
If the job is not core IP, stop paying for seats and pilot an outcome-based capability instead.
2. Where is our failure demand?
Audit services and BPO spend that exists to fix problems your product created.
Do not automate the repair crew. Eliminate the root cause.
3. What department exists only because we cannot trust autonomy yet?
Identify exception-heavy functions.
Build an autonomy roadmap: instrument exceptions, define QA gates, and reduce exception rate before scaling.
Eliminate mess-for-less and build capability without complexity.
Do not build a better firm. Build an atomic one.