Aamir Jawaid

The thing about the U.S. and Israel is that both of their leaders are trying to stay out of prison.

Schedule

Okay fine I’ll learn about jj

Missed opportunity to call it onlyflans

probably the realest thing I ever wrote

I don't think LLMs make work easier. They make work faster, and that is "better" for companies and give workers an approximation of ease bc of the apparent lack of labour. You're still accountable for the output, which makes you a details reviewer, which I often find harder w/o the original labour.

Love this post. It’s gentle and realistic when it comes to adopting ai and comes from such a great pragmatic pov. Thank you for sharing!

I like rust style success/failure result types. Similar to www.npmjs.com/package/ts-r...

How do you make sure snippets in docs are tested? Or do you avoid putting snippets in docs and just point to full samples?

The bit about agents.md being short and a map is really good!

This is insane.

No I don’t care about your ai workflow. Thx

python tooling is notoriously bad. not sure why tbh.

uv has been a breath of fresh air. does most of the heavy lifting for you

In theory, family visiting after the birth of a baby is helpful. It should give us time and space to focus on the baby. But sometimes family visiting can add way more burden to the already hectic initial few weeks.

I feel like a lot of people think about time wrong. Been bugging me since I talked about side projects the other day and was complaining about it.

I had to remind myself that at no point in my adult life has time been given to me, or materialised out of nowhere. You have to actively create time.

3-panel comic from Aram J French from mandatoryrollerscoaster.com: First panel: Grey hoodie guy talking to brown sweater guy: "We invented a robot that answers questions." Second panel: Grey hoodie guy (standing next to robot): "We just have to feed it 10 baby giraffes a day." Third panel: Brown sweater guy: "But it answers the questions correctly?" Grey hoodie guy (making a face as if it was obvious): "Oh my goodness, no. No no no no no"

Neutral then angry wojak

7yo: Dad can I open my new Lego set?

Me: I want you to finish the one you're working on first

7yo: so you always complete a project before starting a new one?

Me:

Being a parent of young kids has made me significantly more empathetic too

We use the #huckleberry app for tracking feeds, diapers, sleep etc. It’s one of the most well thought out apps I have ever used. Every bit of it is so intuitive.

I really like this idea. I never articulated what I felt with having this much decision fatigue as Claude burnout but that’s exactly what it is!

That’s wild

Moto: Containers, not lids

I guess the key question is that if apps have significantly different data requirements, it used to be up to the apps to deal with that. But now if your pds sucks or if you’re on a lower plan on your paid pds, you might not be eligible to use the app, right?

I guess this implies copies of the data being made by the app view. Which sounds okay, but then the app view has to do the work of staying up to date with the pds, right?

Ah gotcha. In practice how does this decoupling work? App Views make a copy of the blob to serve?

Currently YouTube knows its requirements and handles this itself

Right but even data requirements for certain apps is way heavier. If we say that the pds now needs to host videos for ppl to be able to onboard onto at-proto-YouTube, then ppl need to find a pds that can do it or not be able to join the app right?

Generally apps take this responsibility of scale on too. Now is it the job of the pds? Also doesn’t that also mean that requirements for PDSs is kinda open ended? How do apps/pds solve for this potential disconnect?

Helpful piece! But something keeps bothering me about this. I know this promotes the idea that you can host your data anywhere but different types of apps can have significantly different data requirements right? Like YouTube has significantly different data requirements than say X…

Yeah plan mode only really works if you have a very clear idea of what you’re building. For small features of minor additions this is easy. For complex stuff I still prefer the “don’t code just talk” prompt to weigh my options and iteratively build out what I want.