I love your guidance on generating the spec doc. I tried it out this morning and was surprised at how far I could get with Cursor in the first hour. Your guidance also made it easier to know when to create agentic rules because each step of the prompt instructions made it easy to "eject" prompts for cursor rules.
One tweak my GPT made was to incorporate more of the test writing into the process rather than treating it as something after the fact. I imagine that spending more effort on the initial prompts could optimize this.
The use cases you list have a surprisingly low adoption rate. The best use case (stack trace analysis) was only reported by 30% of developers, and the range drops all the way to 10%. It's almost like the chart should be flipped to show use cases developers haven't adopted yet because it shows how 70-90% of developers still need help adopting AI for these situations. How were these use cases chosen?
We did some related research on Dev Interrupted recently. Here are some of the early results:
- The most common use cases where devs are working with AI are writing code (75%), writing tests (68%), and writing docs (53%).
- The most commonly reported use cases where AI does the entire task autonomously are PR descriptions (25%), writing tests (22%), and writing docs (18%)
- 85% of devs are still managing project management tasks entirely themselves (creating tasks, prioritizing work, defining task requirements). There is still a lot of room to grow upstream from writing code.
I love your guidance on generating the spec doc. I tried it out this morning and was surprised at how far I could get with Cursor in the first hour. Your guidance also made it easier to know when to create agentic rules because each step of the prompt instructions made it easy to "eject" prompts for cursor rules.
One tweak my GPT made was to incorporate more of the test writing into the process rather than treating it as something after the fact. I imagine that spending more effort on the initial prompts could optimize this.
The use cases you list have a surprisingly low adoption rate. The best use case (stack trace analysis) was only reported by 30% of developers, and the range drops all the way to 10%. It's almost like the chart should be flipped to show use cases developers haven't adopted yet because it shows how 70-90% of developers still need help adopting AI for these situations. How were these use cases chosen?
We did some related research on Dev Interrupted recently. Here are some of the early results:
- The most common use cases where devs are working with AI are writing code (75%), writing tests (68%), and writing docs (53%).
- The most commonly reported use cases where AI does the entire task autonomously are PR descriptions (25%), writing tests (22%), and writing docs (18%)
- 85% of devs are still managing project management tasks entirely themselves (creating tasks, prioritizing work, defining task requirements). There is still a lot of room to grow upstream from writing code.
You can read about our experiment here: https://devinterrupted.substack.com/p/the-matrix-that-makes-your-ai-strategy