top of page

API-first organizations will require Ops to rewrite everything

  • Writer: Irina M
    Irina M
  • May 7
  • 3 min read

You can outsource your thinking. You can't outsource your understanding.


There's a moment in Andrej Karpathy's recent talk where he describes knowledge moving from text documents to APIs. I wrote 3 words in my notes: this is ops.


At first it sounds like an engineering problem. Collect all your knowledge sources and turn them into APIs. Until you realize most people on your team have never read a command in their life. They never needed to.


I'll give you an example.


Onboarding documentation today is a Google Doc written in plain English. In an API-first world, you define the structure first (schema/workflow/logic) and the text doc becomes a byproduct of that. Agents will be set up to understand everything happening in the company. But regular people still need to read it. Some commands are intuitive. Many are not.


So the question becomes: do we retrain everyone to read API docs, or do we build a human layer on top?


The second path is more realistic. Retraining entire industries at once is slow, expensive, and unlikely to happen fast enough to keep up with how quickly API-first architecture is being adopted. You will need someone to bridge that gap immediately. That someone is ops, and the bridge they build is the human layer.


What the human layer actually looks like


  1. You take an API (a structured set of commands that a system understands) and you translate it into a document a person can read, verify, and act on.

  2. That translation can happen 2 ways. Someone does it manually and an AI reviewer checks it for accuracy. Or it gets auto-generated and a human verifies the output before it goes anywhere near a real process.


Either way, someone owns that layer.


Why regular people will learn to read commands and why it will take 2-4 years


People are already learning to read commands without realizing it. When someone vibecodes, they see code. When they set up an agent, they see commands. When they do integrations in Canva, they are reading something that looks a lot like structured logic. The concept of a command is slowly stopping feeling foreign.


The comparison to prompting makes this clear. When prompting emerged it felt technical. Using specific structures, assigning roles, writing "imagine you are a marketer with 10 years of experience"-none of that was obvious. People who knew nothing about AI took several years to get fluent. And prompting is actually easier than reading API commands because you are still using natural language, just more precisely.


Reading commands is a different skill but exposure is already happening whether people seek it out or not.


The wildest version of this (and I genuinely believe it) is that in a few years your grandmother will know what a basic command does. Because it will be everywhere.


My speculative guess is 2-4 years before basic command literacy becomes unremarkable in professional settings.


What this means for ops right now


While that shift is happening, someone has to catch edge cases. Someone has to own the iteration loops when the system breaks in production (and, yes, it will break).


Someone has to know when the output is wrong even when the system is certain it is not.

That someone has always been ops.


Ops has matured into a hybrid of product thinking, system design, and quality control.

 
 
 

Comments


bottom of page