Błażej Nowakowski

There is insane demand for people who can understand and explain technology in a compelling way.

AI is DX 2.0 Many of the things I considered "state of the art" in developer experience are 10x easier with AI. And... it's only going to get better, so it's worth starting to think about tasteful ways to leverage AI. I'll give you some examples I've been exploring. One of the marks of a company that has great DX is that they have wonderful docs. So, how do you actually write great docs? It's not just about having good writers (or training them). But building a system to allow hundreds of engineers to write or edit docs without it turning into a huge mess. Style guides 👏 But, just like how everyone hates linters, people who write docs hate copy linters. e.g. I'm trying to fix this PR and it's complaining because I started this sentence with "But". Come on now. They serve a good purpose though! I want to enable engineers to help make the docs better, not to funnel it through one small team, while ensuring our docs have one common voice. So let's ask ourselves, can AI make this better? Rather than some nasty pre-commit hook, you can have a CI script that looks at your git diff. Based on the diff, you can ask an AI model to apply your style guide, and output JSON¹. This JSON can match the format GitHub expects to apply suggestions to pull requests. Now, rather than linters telling me I'm holding it wrong, the AI comments on my PRs with suggestions I can approve or deny. And because AI models have broad world knowledge (especially as it relates to technical terms), it's a helluva lot better than a hardcoded list of words to allow or disallow. Here's another example: customer feedback. If you have the privilege of making something that people want, they're going to share with you how they feel about the product. So, what do you do all that feedback? How do you separate the signal from the noise? Sure, you could probably buy a SaaS product that does this. But with AI, you can very quickly go from a firehose of data, to extracting out the most relevant pieces of feedback and piping it to a Slack channel or other destination. One final example. Many docs have quickstarts or tutorials you can follow to get up-and-running with the product. But you know what's better than following a tutorial? If the entire thing can be automated. The developer just wants to see something working. They're in problem solving mode. If you can get them from idea to working app faster, the more likely they're going to have a good time. To this point, we've been adding buttons to "Open in v0" to our docs code snippets. Why listen to me go on about how to set up the feature when you can press a button and try it live? Then, remix it and make it your own. Magical. These are things I'm actively building where AI is making it much easier. It's time to revisit your priors. Curious if others have similar anecdotes. —— ¹: Bonus: if you use the AI SDK with `generateObject`, this is so easy. Plus JSON mode with OpenAI and you're cooking.