The job is no longer writing code. It’s writing intent. We are seeing that frontier developers hand-write less than 1% of the code they produce. The rest comes from agents that they instruct, steer, and review.
The work becomes two things: defining what “done” looks like (requirements, constraints, acceptance criteria) and verifying that the output meets your intent.
The discipline of writing clear intent isn’t new, but it’s no longer optional. Describe the behavior you want, the edge cases that matter, and how to verify correctness. For example, “Add auth to this API” is not clear intent; specifying which roles can access which endpoints, what happens on token expiry, and how to test unauthorized access is. Then let the agent write the code.
Where you once spent most of your time writing code, you now spend it on intent, direction, and targeted review of what agents produce.
The job is no longer writing code. It’s writing intent. We are seeing that frontier developers hand-write less than 1% of the code they produce. The rest comes from agents that they instruct, steer, and review.
The work becomes two things: defining what “done” looks like (requirements, constraints, acceptance criteria) and verifying that the output meets your intent.
The discipline of writing clear intent isn’t new, but it’s no longer optional. Describe the behavior you want, the edge cases that matter, and how to verify correctness. For example, “Add auth to this API” is not clear intent; specifying which roles can access which endpoints, what happens on token expiry, and how to test unauthorized access is. Then let the agent write the code.
Where you once spent most of your time writing code, you now spend it on intent, direction, and targeted review of what agents produce.