Software development is shifting away from direct code entry in an IDE toward prompting an LLM and reviewing what it produces. The programmer's role becomes oversight: humans remain accountable for what the software does and how it works, but apply a different skill set to build it.

Where agentic engineering sits

Agentic engineering differs from Vibe Coding in how the code is treated. In vibe coding, humans do not look at the code and treat its existence as irrelevant. In agentic engineering they stay concerned with it and often review it in detail. It also differs from using LLMs as sophisticated code completion inside an IDE, where the model merely assists code that a person is writing.

How the workflow runs today

Current agentic tools operate in a terminal environment. The programmer issues prompts, which often carry saved documents describing guidelines the LLM should follow. The LLM then works on the source tree directly: creating and modifying files, running the code, evaluating test results, and continuing development, frequently for extended periods. When it finishes, humans assess the work through code review, examination of test results, and review of output from other sensors.

What programmers need to develop

This change raises broad questions about which activities programmers will perform and which skills they will need. At this point harness engineering — work on the guides and sensors around the LLM — appears central. Understanding the domain also grows in importance, as does collaboration with users and customers to iteratively define and build the product.