We've actually seen something very similar on our team.
We had an idea for a feature that would normally go through the usual sequence: write the spec, review, plan the work, and then someone starts coding.
One engineer did almost all of that himself. He drafted the PRD with AI and then fed it into a coding agent, which came back with an implementation plan.
Writing the code ended up being the smallest part of the work. The attention went to reviewing the plan and deciding what was worth keeping .
Is amazing when we see this kind of workflow working. Which means the future will be more about getting in alignment and experimenting with features. than writing the code itself.
I truly believe that Vibe Engineering :) is the future of coding.
Interesting. My view was more aligned with CRUD/Enterprise applications. Regarding embedded systems and IoT, it goes outside of my experience since I didn't work with these requirements in a project.
However, if it is not the case today, I would bet it will likely be in the future. For CRUD, the technology doesn't matter anymore, even though AI uses very well-known technologies. Which means that for a system that requires lean resource consumption (IoT for example), AI can still develop these systems. If it is not doing it today efficiently, I would bet it would be a matter of time.
For a security-hardened system, there may be use cases that require full agency of human or developer code generation. However, the trend will be for the developer to have an oversight of the architecture and implementation, rather than code generation.
Nevertheless, I mention "for the vast majority of business applications," giving a space for very challenging implementations that might require human code generation. Even though IT Security is an area where AI is already helping immensely in system design and its implementations.
What is your perspective? What is your view on the AI use/vibe on coding?
We've actually seen something very similar on our team.
We had an idea for a feature that would normally go through the usual sequence: write the spec, review, plan the work, and then someone starts coding.
One engineer did almost all of that himself. He drafted the PRD with AI and then fed it into a coding agent, which came back with an implementation plan.
Writing the code ended up being the smallest part of the work. The attention went to reviewing the plan and deciding what was worth keeping .
Is amazing when we see this kind of workflow working. Which means the future will be more about getting in alignment and experimenting with features. than writing the code itself.
I truly believe that Vibe Engineering :) is the future of coding.
What about embedded systems? IoT? Security-hardened systems?
Interesting. My view was more aligned with CRUD/Enterprise applications. Regarding embedded systems and IoT, it goes outside of my experience since I didn't work with these requirements in a project.
However, if it is not the case today, I would bet it will likely be in the future. For CRUD, the technology doesn't matter anymore, even though AI uses very well-known technologies. Which means that for a system that requires lean resource consumption (IoT for example), AI can still develop these systems. If it is not doing it today efficiently, I would bet it would be a matter of time.
For a security-hardened system, there may be use cases that require full agency of human or developer code generation. However, the trend will be for the developer to have an oversight of the architecture and implementation, rather than code generation.
Nevertheless, I mention "for the vast majority of business applications," giving a space for very challenging implementations that might require human code generation. Even though IT Security is an area where AI is already helping immensely in system design and its implementations.
What is your perspective? What is your view on the AI use/vibe on coding?
Thanks for your extensive statement! Like you said, it’s not black and white and there is definitely the trend you describe.
We use vibe-coded code after human review, but we stick to humanly designed architecture.