Andrew Atiya

Multiplayer AI is the future, rooted in the past. I have never been more excited about building. I want everyone to have this. My thoughts, here: x.com/atiya_ag/statu…

@tednotlasso Yes. I don’t think we’ve fully confronted the harms of turning people into slop cannons. Startling how quickly the information environment decays. Notable that in this corner of twitter we’re always checking posts with pangram.

“Information is cheap to generate but it costs everyone’s time and energy to review.”

@paulg General pattern when evaluating the new thing is we wildly overestimate how good we were at the old thing.

Before vibe coding became a thing, programming was already evolving in that direction. It already increasingly consisted of installing and configuring stuff other people wrote, without reading the source.

What is ancient will be new again. x.com/atiya_ag/statu…

@mikeldking Great practical advice. Resonates with my own experience. One of the quiet revolutions happening now is how fast and portable devops has gotten.

Over the past 6 months we've maniacally prepped our repos to be coding agent friendly. Here are some things that worked. 1. Make CI blazing fast. Use every Rust, Zig, or Go ported tool that lets agents verify their work. This means UV, oxfmt, Typescript 7. Move integration tests to post merge hooks. 2. Trigger coding agents automatically based on triage labels. A coding agent should setup a proof of concept or repro steps automatically so an engineer can pick up the issue seamlessly. 3. Setup crons for things devs hate doing. Setup agents to fill SDK gaps, skill tuning, filling in critical regression checks. 4. Give agents ways to prove their work. Add screenshotting skills, agent-browswer, cloud storage for storing assets. 5. Make it possible to hermetically deploy your app, preferably multiple at a time. If a coding agent can deploy the app locally, the faster it can work. 6. Give the agents realistic production "simulation" data. Agents will work much better when they are working against data that looks like how your users use the product.

@AnikaSomaia voice dictated "make google" on my morning walk it then invented MapReduce

@atiya_ag woah one-shot too?

new genre of tweet: repost something you built a year ago and caption it "made with fable in 10 minutes." incredible economics, you get credit for the work and credit for not doing the work

Recurring lesson from being an AI-pilled startup: AI is very helpful for doing work AI is very bad at doing thinking If allowed to own the latter, it eventually chokes on its own effluence

@gerstenzang Yeah it’s wild anyone thinks this is going to work. If it did aws would sell crm, email, ride sharing etc etc etc

It is highly unlikely that the foundation companies will be successful in all - or any - of these categories. As always, the hard part isn't the code: it's the workflows, the go-to-market, the servicing. I don't think there has ever been a platform company that has successfully won a vertical category, and for good reason. So why are they saying they are going to all these things? 1/ to "lick the cookie" devblogs.microsoft.com/oldnewthing/20… 2/ a fit of exuerbance 3/ a real need to show they can do something that can't be commodified

@satyanadella Great to see this question getting attention. Is the answer obvious? Your company's real expertise is its people. Always has been. If a company's people don't know something, the company doesn't know it either. Doesn't matter where else that knowledge resides.

@eve_bouff Her literary interview series is incredible. She almost made Ocean Vuong cry.

this is so fucking hot. dua lipa is the renaissance humanist we didn't know we needed

@gordon_cassie Anyone who allows this, in law or any other expertise heavy business, is set for a rude awakening.

Super interesting thread in r/biglaw. Many people chiming in with similar experiences.

Update. No it doesn’t! Actually caching is quite good. Just very hard to configure correctly

Particularly odd, doing the things that are supposed to improve cache hits have actually made it worse.

gpt 5.6 seems to have very different caching behavior than 5.5/5.4 Luna is averaging 2.1x more expensive than gpt-5.4-mini in our production workloads due to much lower cache hit rate. Still figuring all the new knobs for this but I wonder if this is related to usage limit gobbling everyone is complaining about.

@sama Create a bigger pie and everyone wins. Bigger ecosystem, more jobs, more opportunities, more benefits to the economy and society. The negative sum game case never made sense to me. I’m glad to see it fading away.

so far at least, i'm pretty sure AI has been net job-creating. this was not what i expected--although i was much less pessimistic than others, i thought by this level of capability we'd have seen some impact. it is possible this direction keeps going!

@ImLunaHey @t3dotcodes 🥳

@atiya_ag @t3dotcodes seems to also be usable on their website so im inclined to believe its in the $20/m sub.

wtf have i been missing. codex is sooooo much faster than claude.

@ImLunaHey @t3dotcodes 🤷‍♂️ if it works it works? You can also try the codex desktop app too. It's the only one I actually prefer over Zed

@atiya_ag yeah seems to be using sol. unless @t3dotcodes is showing it incorrectly.

@ImLunaHey You are? I dont think its included but I could be wrong. I'm that guy who exhausts limits on the $200/m plan so my info might be out of date

@atiya_ag ah.. im using sol on the $20/m plan?

@ImLunaHey Yeah the $20/m plan is kind of like the free hit a drug dealer hands out. Especially since it doesn't include sol.

@atiya_ag honestly imma give codex a day on the $20/m plan and if it keeps up with this ill probably cancel my $200/m plan with claude tomorrow.