Experimenting with AI in DevRel
Using AI tools like ChatGPT and GitHub Copilot to enhance DevRel work, build inspiring code samples, and enhance developer experience.
At Nylas, we have a company-wide mandate to get skilled at using ChatGPT by using it twice daily: once for a work task, and once for a personal task.
I'm personally way beyond that minimum usage already. I'm encouraging my DevRel team to push further as well. But I don't want to just say "Hey, use the AI more".
I offered the team a simple thought framework, which I'll share in this post.
Unpacking where AI could be useful in DevRel
The "framework" is a simple list of questions, which I offered the DevRel pros on my team to illuminate areas where they might apply AI to their work.
The questions, in what I currently believe is an increasing order of complexity, risk, and opportunity, are:
- How can AI enhance the way you work?
- How can developers use third-party AI to enhance their integrations with our APIs?
- How can AI enhance the developer experience we offer?
- How can AI enhance our product?
This is where the framework stopped in terms of how I positioned it to the team. It's too early to be prescriptive and I don't want to crowd out creativity with my own still-forming thoughts.
But I do want to share some ideas somewhere. So here are a few quick thoughts on each:
Areas for AI enhancement | Avenues to explore |
---|---|
Your work | Drafting blog posts, generating docs, code completion for sample code, etc |
Developer use cases | "Our APIs + AI APIs", probably through a CRUD lens to start |
Developer experience | GitHub Copilot for Docs, Discourse integration, use case recommendations, SDK and dev tool enhancements, etc |
Product features | Common use cases that are tedious to build, the recently-possible-thanks-to-AI, etc |
For the DevRel team, I encourage us to make a lot of space for the first three questions.
In some ways, the forth question—product features—seems easier to have ideas about but there's a lot of rigor that should go into the product validation and the engineering implementation that is outside of our mandate as a DevRel team. So yes, ideas for product features are welcome, but I want us to challenge ourselves on the areas we can directly impact. (Plus, with the current reactionary "let's toss ChatGPT into it" product fad, I wonder a lot about early adopter syndrome and user fatigue.)
Examples from my own work
This is obviously early days still, but here are some examples I've been playing with for the first three questions above.
- My work: I just used ChatGPT to help me draft a new version of our post-signup email nurture program. The conversation produced an 18-page PDF when exported. I'm not sure I can say doing things this way saved me time, but it did help me round out my own thoughts.
- Developer use cases: I pushed a sample repo to GitHub showing how to use the Nylas Email API and the OpenAI API to find important emails in your inbox. This also involved some of #1 just above, since I was using GitHub Copilot while writing the code.
- Developer experience: I'm currently experimenting with what AI could do for our Docs. I don't have access to GitHub Copilot for Docs yet so I'm using this time to form my own opinions. Feeding GPT-4 an OpenAPI spec and then talking to it is wildly fun.
I'm using the current AI offerings for other things as well, but I think the above are the most clear-cut as examples.
Am I too hyped about AI in DevRel?
I say no: I'm clear-eyed about what AI currently is and what it isn't. It isn't a solution for everything but it's a great thought partner at the very least.
For all of the above examples from my own work, the AI is contributing to the process, but I am the final decision-maker, and am ultimately responsible for the final creation.
My approach can be summed up as "Experiment aggressively in private; ship to users (developers) with care and consideration".