Agentic Programming

Increasingly software developers are not typing code into their IDEs. Instead they prompt an LLM to do so, then review the results. This is a profound change to the nature of programming, where humans oversee LLM agents who generate the code. Humans are still responsible for what the software does and how it works, but use different skills to create their products.

I like to distinguish agentic programming from Vibe Coding. With vibe coding humans don't look at the code, indeed they forget that it even exists, while with agentic programming they are concerned with the code, often giving it detailed review. Agentic programming is also distinct from using LLMs as a sophisticated code completion mechanism, where LLMs participate in writing code inside an IDE environment.

This way of working hasn't got a widely established name, but the “agentic” adjective is common, used as “agentic programming”, “agentic (software) engineering, “agentic coding”, etc. (Currently I'm inclined to go with “agentic programming”.)

At the time of writing, agentic tools work in a terminal environment, the programmer issues prompts, which often incorporate saved documents explaining guidelines that the LLM should follow. The LLM then manipulates the source tree directly, creating and modifying files, running the code, evaluating the results of tests, and continuing development work, often for extended periods. Once done humans evaluate the agents work, doing code review, examining test results, and reviewing outputs from other sensors.

This shift in the nature of programming raises extensive questions about what kinds of activities programmers need in the future and what skills we require. At this point harness engineering, focusing on working on the guides and sensors around the LLM seem central. Additionally this raises the importance of programmers understanding the domain they are working with, collaborating with users and customers to iteratively define and build their product.