Blogmark

The math has shifted, in some cases, significantly.

Coding agents dramatically drop the cost of typing code into the computer, which disrupts so many of our existing personal and organizational intuitions about which trade-offs make sense.

That prototype, that bug fix, that nice-to-have -- those things that tend to get pushed off over and over because more urgent things are taking your time -- are suddenly viable in a lot of cases with an LLM coding agent. Equipped with a paragraph of detail and using something like Cursor or Claude Code, we can hand off these back-burner tasks, iterate on the idea in minutes instead of hours, and have at the very least an MVP if not a ready-to-ship implementation. All with a relatively minimal interruption to the main task at hand.

While we're trying to catch up to what is possible here, all the different processes our organizations and engineering teams have surrounding how we go from feature idea to shipped and supported implementation of said feature have some catching up to do as well. In some cases, these processes are meant to slow things down, as a way of managing risk. These processes are in opposition to the mandate to using AI agents to ship faster. What are organizations going to do with this contradiction? That's an open question.

I also like that as part of this series Simon is putting forth the term Agentic Engineering as the other side of the spectrum from vibe coding when it comes to using LLMs to write code.