mike suarez
@msuarz.bsky.social
45 followers 120 following 360 posts
natural born programmer
Posts Media Videos Starter Packs
msuarz.bsky.social
i trust tests and clean code … both r really east to check … getting clean code takes effort… but get test green n the drill sergent with coding rules
msuarz.bsky.social
can anybody in github have some decency and add test reports to gha pls … i cannot convince the world that gha is a stupid ci system … could github pls pretty pls try a little bit!!!
msuarz.bsky.social
man i could not put thru the nostalgia bs of the first part ... too soon for that ... given current nostalgic societal situation of making things great like they were :( f that
msuarz.bsky.social
idk i keep hearing that claude code secret sauce is unix philosophy
msuarz.bsky.social
i'll take u on this one @pragprog.com just too funny
msuarz.bsky.social
i just punned the dude ... llmao
msuarz.bsky.social
well crypto is definitely no fun ... at least we enjoying the burning of the amazon now
msuarz.bsky.social
no hear me out ... the model of today makes money tomorrow ... but u need to invest in it today ... then tomorrow u have payoff but u have to invest in the after tomorrow ... so they are always making money and in debt .. i swear that's the explanation lmao
msuarz.bsky.social
tomorrow for vibe coding im gonna take a stab at @emilybache.com lift kata … my spin is with react snaps for approvals and the dude at the wheel … here’s a sample test github.com/limadelic/el... and the schnaps github.com/limadelic/el...
msuarz.bsky.social
ai is not mind reader it needs instructions .. so code becomes instructions which u need to manage to avoid mixing concepts and having the llm focus on problem size it can do reliably … the agentic loop thing is the back and forth to achieve the goal of your method call
msuarz.bsky.social
i got this working with tests and all that … an agent is a class … when running is an object … it receives messages and responds to them just like calling a method on a class … the llm is the brain … agent instructions r in plain english
msuarz.bsky.social
same as now u work on a story and change your agents instructions or add new agents
msuarz.bsky.social
my take on it is classes become agents .. thats a size of agents quite diff from ur reg agent .. then those tiny agents need little instructions just declarative … llm executes the instructions
msuarz.bsky.social
this switch 90/10 to 30/70 is more interesting :) did u get the names right u saying more codex now right? that itself is more newsworthy cos claude code seemed to b king of the hill for a little while
msuarz.bsky.social
cmon dude it writes 100% of mine whatevs ... but does it do it by itself?
msuarz.bsky.social
still happier with codex over cc ?
Reposted by mike suarez
simonwillison.net
I'm ready to accept a definition of "agent" that I think is widely-enough agreed upon to be useful, at least in the AI engineering space:

An LLM agent runs tools in a loop to achieve a goal

(I've been dismissing the term as hopelessly ambiguous for years) simonwillison.net/2025/Sep/18/...
I think “agent” may finally have a widely enough agreed upon definition to be useful jargon now
I’ve noticed something interesting over the past few weeks: I’ve started using the term “agent” in conversations where I don’t feel the need to then define it, roll my eyes …
simonwillison.net
msuarz.bsky.social
what that one is months old … hardness… agent hardness is the new one
msuarz.bsky.social
i dont say enuf how much i hate TS