Vibe coding (or agentic engineering) is really just a form of automation – it automates the process of writing code. As with other forms of automations, its effectiveness depends on its surrounding. For example, suppose that you have a fancy IoT coffee machine that can be programmed to make a coffee automatically every morning right before you wake up. It can’t be very useful if it is not connected to potable water and has fresh beans in its reservoir.
In the past, we can often get away with a few bad habits in our software engineering lifecycle, but they now matter more than ever for organizations that want to adopt agentic engineering. Here are some examples:
- Version control: When you are working mostly alone, version control might not be that useful. When working with an AI agent? Good luck.
- Automated tests: For smaller projects, we can get away doing manual QA after every change. However, that will quickly become the bottleneck with agentic engineering.
- Documentation: AI agents do not have access to institutional knowledge in our head. They need written documentation to understand past decisions.
It goes beyond just the practice of writing code and it extends also to the more abstract, such as the organizational culture. Does your organization value experimentation? What about cross-functional collaboration? Are employees encouraged to challenge the status quo?
In the coming few years, we will hear more leaders pushing for “AI transformation” in their organizations. They should know that in order for that effort to be successful, it is not enough to just give everyone an OpenAI or Anthropic or Gemini subscription. They need to also think about everything else.