Labor as code

Labor as Code™ has four steps and they run continuously. Decompose a role into the tasks it actually contains. Make those tasks visible, with inputs, outputs, and timing. Automate or assist the tasks that meet the preconditions. Measure what happened and repeat.

The chain runs from labor to tasks to workflows to code to AI-assisted execution. Most organizations attempt to jump from labor straight to AI-assisted execution, which is why the result is usually a demonstration rather than a change in how work gets done.

A care coordinator does not do care coordination. They chase records, verify eligibility, place calls that go unanswered, re-place them, book appointments, send instructions, confirm attendance, and escalate the ones that stall. Those are different tasks with different volumes, different failure modes, and different suitability for automation.

A registered nurse, a clinical pharmacist, and a medical assistant decompose the same way. Some of what each does is judgment under uncertainty with a licence attached. Much of the rest is retrieval, drafting, checking, and chasing. The distinction is not visible at the level of the job title, which is exactly the problem.

Decomposition is the expensive step and the one that gets skipped. It is also the one that produces the value even if no automation follows, because a task list with volumes and failure rates is already a management artifact the organization did not have.

The Zero Neo interest here is narrow. If AI has made configuration cheap, the binding constraint moves to understanding the work well enough to configure anything. Labor as code is the method for getting that understanding, and its cost is now the number worth watching.