Enterprise Operator. AI for companies that matter.
Joined February 2025
- Tweets837
- Following402
- Followers89
- Likes14.8K
Ryan Marsh retweeted
Back when Linux sucked we all knew it sucked and everyone was quick to admit it was a hobby, a yak to shave. Then it got just a little better, barely useable, and tons of idiots ran around trying to get people to switch from actual functioning desktop operating systems and that’s what the small local model people remind me of.
Ryan Marsh retweeted
working for yourself is addictive in a way most people don’t anticipate & the stress that comes with it is genuinely difficult to manage. when you’re the principal, nothing happens without you. you set the tone, the structure, the culture, & the motivation. you are the foundational layer.
most ppl have never had to operate this way cuz they’ve always been embedded in a system & told what to do, when to do it, & what matters. aka the shawshank lifestyle. the human mind runs surprisingly well on prescription. strip that away & you discover pretty quickly whether you actually have an interior architecture or just a talent for compliance.
that’s what makes it so damn dichotomous. the freedom is so fucking real, but so is the weight. all i’m saying is that most ppl have never had to locate their own gravity before.
Ryan Marsh retweeted
Over my career I've seen a _lot_ of attempted complete rewrites of existing software, and two things have stuck out to me:
- The overwhelming majority of rewrites failed
- Of the small minority I saw succeed, almost 100% were done by the same team who wrote the original software.
Ryan Marsh retweeted
may have hit a nerve here...it all started with trying to understand how production AI systems actually work
FOLKS - I've tried every agent framework out there and talked to many strong founders building impressive things with AI
BUT I was surprised to find that most successful AI systems aren't following the "here's your prompt, here's a bag of tools" pattern
► they're mostly just well-engineered software with LLM capabilities integrated at key points
The companies shipping high-quality AI aren't building monolithic "agents" from scratch
► they're incorporating small, focused llm loops that do one thing well
so I set out to document the principles for building production-grade LLM applications in "12-factor-Agents"
We were on the front page of HN all day on wednesday, with great discussion from the community
check it out below 👇 and let me know what factors you think are missing
Ryan Marsh retweeted
LLM vendors should be the ones shipping eval tools & frameworks. If customers pick models based on evals, vendors need to know why. What am I missing here?
Ryan Marsh retweeted
“MCP = GraphQL for LLM” is a good mental model.
Is Apollo making money? Yah, but where’s most of the value? In the systems where the actual data resides, and in the clients using the data.
