Post

No I've not dug into those at all yet - they do feel a bit more accessible now than they used to, but I'll still need to spend a bunch of time to get up to speed on them

it's the Truman Show approach to testing. i love when engineers go-to-town with LLMs it blows your mind! (yeah tradeoff is it blows the wallet too..)

www.strongdm.com A time machine to the 90s web by the looks of it

Mind blown by the 'no human-written code' idea. How do you think this will change the dev landscape? 🤔

Automating QA is a thing on my to-do list. I've got quite far in the thinking and, for pure speed practicality, I'm looking at systems that minimise agent use. IE, they run many times and turn to the agent when something unexpected happens.

That sounds sustainable.

lol thanks for the update because i literally closed the article this morning when i read that line. i’ll go back and check it out now.

I think it’s a pretty good marketing/filtering tactic. A analogous to how @steveklabnik.com interpreted the original Gas Town post steveklabnik.com/writing/how-...

Without divulging the details since I really want to stay anon here, the task was implementing an AST for a subset of Lua with a lot of wrapped infra code for AWS+ UI. I didn't want to offload all the parsing to the LLM because it's a tractable problem.

I think a lot of these claims are a pissing contest. Reminds me of people bragging about their sex life in high school despite not having one.

1000/day is 365k/year, or about 1 fte at ~180k/yr salary with the simplest adjustment for company costs to hire someone. If you convert token costs to subscriptions, there’s around a 10x subsidy. So you’re comparing 1000/day in token spend to 30 max plans. Your 1max plan idea is 1/30 of a 180k dev

The $1k/day number reads differently depending on whether you think token costs will keep dropping or plateau. If inference gets 10x cheaper in 18 months, today's burn rate is actually buying future-proofed institutional muscle memory. The bet is on the trajectory, not the snapshot.

I think a very (excessively?) generous reading of that is $1000/engineer/day *at the current cost per token which will eventually come down*

Yeah, I didn't get that amount. That seems insane to me. "I mean, it's one token, Michael? What could it cost? $10?"

Perhaps a “bunch of goto statements” is simply the next viable language pattern — one that LLMs inherited to move beyond human coding style. As long as it works, who will really care? Maybe only a few old-school nerds like us 😉

The folks at Truong DM are great! I’ve been working with them on Cedar (you may find that interesting too, especially given the AI friendly, automated reasoning) and was unaware of this

Thanks for that!

@unroll.skywriter.blue unroll please

We got so good at this (which means we are so so far behind people like Anthropic) that we were 3-4 weeks behind on PRs. Reminds me of MBA use cases about IBM when Agile came out and QA couldn’t keep up with the new speed

This approach is fascinating. Magic sauce seems to be the agentic scenario validations

I'd love to get more detail on how that works too, their write-up is the most I've seen so far