I really loved sqlc's approach, so I'm glad there's another option like it for Rust!

I recommend Amp if you can afford passthrough pricing! I appreciate that they choose the best models for the job and I don't have to reconsider which models to switch to for different tasks every month

4. If we enjoy working together through the trial and aligned your value to your desired compensation

Only through the experience of a trial can we actually understand if the compensation you're seeking is aligned with your impact. And it confirms for you whether you enjoy working here with us.

3. If we work well together

Working together is natural and enjoyable as we work through an interesting technical design problem. You balance building with thoughtful API design and can talk through how it might work well or not for AI agents to maintain

2. If we enjoy chatting about work

It feels fun and effortless to chat. We can nerd out about similar things. I naturally learn something interesting through our conversation that I'll want to apply immediately.

1. If you appear to be good.

You are either a prolific product builder with experiences pivoting OR you have previously built something we need help building

I'm working through one of my toughest system designs ever—but I'm genuinely enjoying it. Usually, engineers steer the conversation too deep, while non-technical folks aren’t sure how to navigate the topics to ask great questions.

That's pretty cool honestly

Lovely primer on oklch, it feels like the cheat code to good color picking when you realize oklch and hct (material design) exist

Reasonable change. I have a pretty massive Svelte 5 codebase, but I think there are only a small handful of effect$ instances compared to state$ and derive$.

Most AI eng I see is usually just one step and done. Embed, store, retrieve, and synthesize. I want to see a portfolio where you have added steps in between that benefit the end user. Metaprompting, strategically using scratchpads, memories, tool loops with budgeting, etc

That would be taking novel approaches to solving problems with multiple steps. For example, show me how you wrote one AI to generate intermediary code to be run and used to scrape a website or organize data from multiple systems (this is something happening at Ramp when they import from new banks)

But, I would probably just end up abusing this for very little gain.

There was a cool use case for this with typed link parameters for the Qwik router as well

Cool things are unlocked when TypeScript template literals become more capable of enforcing types. I'm interested in the inference as well, so you can retain the type based on what is written. For example, err`Permission not granted to ${user}` could be `Err<["Permission not granted to", User]>`

Planning through prompting as well?

I learned about this from Stephen King's _On Writing_!

Yeah, that's just a different role at that point. If they come into he team not having a portfolio of AI experience with the intent to "train up" it's high risk to me that it doesn't click for them.

Like hiring a production artist for a role in brand design

can you meet with a customer trying to accomplish a task that could be advised by those documents and design a pipeline that connect the dots to complement their workflow when the task they are accomplishing is completely unique to their IP?