João Fé retweeted
I switched to NixOS a couple years ago and love it. especially now that i can direct my AI to update my nix config.
We will find people defending both sides and claiming each is amazing. But it really doesn’t matter whether you have a monorepo or multiple. What matters is tooling to search your organisation’s source code, docs, etc.
I don’t understand big software companies that are spending dozens or hundreds of thousands dollars in proprietary LLMs per month instead of investing in their own.
They are paying to lose their intellectual property.
I think current AI is a knife and will never be more than a knife, no matter how much you sharpen it.
The reason I have little interest in optimising agents (harnesses) is that my view of AI is one of autonomous self-improvement and continuous learning.
Sadly current AI isn’t like that, so I don’t think it can go far. (It has gone further than I anticipated though,I must say.)
One of the best things about agents is that I can now forget about many of the irrelevant things, and the monorepo vs multi-repo is definitely one of them.
(Of course I also forgot about git.)
who thinks monorepos are good for agents
they absolutely are not, it's a completely wrong pattern for agents, it ruins their ability to navigate history because it removes the project specific container
there's zero benefit from monorepos now that we have agents, only downsides
monorepos were solving a different problem that's no longer even relevant
“But a team of two who always pair-programmed together would not be much of a team.”
Perhaps the Proton CTO should invest in better engineers (humans) than in agents.
Apparently fireworks are illegal in Ireland and yet I’ve heard more fireworks tonight than in all my life before.
Failed country.
Everyday I have good and bad experiences with agents.
Most if not all my good experiences are in the context of searching, basically.
When it comes to decision making and creation it cannot be left alone, and it can get quite counterproductive sometimes.
I guess the “we can fix everything” only works when you are funded a few million dollars in tokens.
with current token prices i wouldn't really say "cost of software is going to zero"
our token cost is double our payroll which means even though we're 25 it's like we're 75
and i feel like our output is similar to a team of 75 pre-ai. we're just able to do it without all the downsides of a larger team
this is still a good thing. we couldn't hire 50 more people. so much overhead and impossible to retain a high bar
I like the fact that Claude Code uses curl to fetch a page when its fetch gets rejected due to some security policy.
I only wish to be as 1% based about everything as Chris is.
You may be a little confused. Programming consists of instructions entered into memory in a DTM, sharing memory space with input data. A procedure determined by the instructions then operates on the data in microscopic detail to produce output.
Everything from one end of this process to the other is deterministic. As a deterministic Turing machine is the only kind of Turing machine that actually exists, this is where you stay put. Whether or not you can follow what's going on inside the computer is irrelevant.