we've taken the first steps in eliminating opencode's ripgrep dependency :)

"x is too dangerous so i am the only one who can be trusted with it" is such a hilariously cliche trope it shows up constantly in history, books, movies and it's the same outcome every time and yet we're watching this happen in real time with AI

raw LLM prices don't mean much but we can look at real world data to see what people actually spend from last 7 days avg cost per session

the whole TUI vs GUI debate no one makes the obvious point why would you want to use something gooey ew gross

did everyone end up adopting .agents/skills convention i wasn't following

wow we're getting a vps vs serverless moment in 2026 bringing back the classics

all these "ai changes everything" threadboys look like they fuck with airpods on

one of the most annoying things in the world is mid level employees at a hot company these mfs think they're such hot shit guarantee they name drop their CEO by first name

you can learn a bit from the ai.​com launch and why the whole "all publicity is good publicity" shit is so wrong - you go to their site - absolutely zero info about their product - fine will try to sign up 1/3

scaling felt solved in software for a while - so many services that solve it for you, so few workloads that needed something special but we're back to figuring out scaling from scratch for inference things feel how they used to 15 years ago, everything constantly overloaded

why did they take so many pictures

opencode zen produces a huge amount of the world's demand for open source models we started routing a portion of kimi k2.5 traffic through openrouter and instantly made it the #1 programing model by a huge margin

we've improved our traffic routing for kimi k2.5 (paid version) on zen so it should now be absurdly fast might hit rate limits here as we scale up but this feels like something different

this is our new ai slop pr policy

i used to have a thousand good ideas and no time to work on them now that i'm a lot more experienced i rarely ever have a good idea

so what's going on with meta ai

just one more markdown bro plz just one more and it'll do what you need

what i learned from this past week is you should launch ai products in texas

are we misunderstanding this? the implication is you can't insert any content that anthropic didn't know to have generated this breaks things like switching models mid session and a dozen other things harnesses rely on i switch between claude and gpt all the time :(

i've been a manager, i'm partially a manager now and let me tell you something it's absolutely nothing like managing a bunch of ai agents

seeing this new agent teams feature in cc i'm kinda wondering if i'm officially old i remember seeing older programmers have no ability to get how the kids were doing things i can feel my brain just being like huhhhh with all this stuff

who do i know that's going to react miami or AI Engineer miami? doing my party again

oh shit oh FUCK

i think a lot more about how things would be with 10x faster LLMs than 10x smarter LLMs

i have so few complaints about the frontier LLMs that idk if i'm going to even be able to tell when they improve

the whole "ai turns on us" risk might be a lot more boring than in the movies even at current intelligence levels if there's a million agents deployed they could get into some dumb loop where they're accidentally DDoSing something important

is it possible to make software that you're proud of

investigating a bun issue - normally i'd use codex for something like this but wanted to see how kimi k2.5 would do - didn't expect much but it looks like it figured it out - and really fast too https://opncd.ai/share/jupHGgzB

openai has a product that is an input box where you could type literally anything into and have a magical experience when you have something this compelling your goal reduces into a simple thing - how do you get every single person in the world to try it 1/4

started work on something complicated - instinct is to go write some code and think through it let me try it differently and use the LLM to create a plan first it just doesn't work when i don't really know what we need to build thinking through by typing is still important

huge personal milestone for me after 15 years as an arch linux user one of our packages has been moved into the official Arch Linux repositories

things don't have to be great to work most food is bad, most software feels terrible when you come across something that feels great it's because they did more than they had to and rejected the tradeoffs everyone told them were real

so we were the only idiots trying to support linux windows and macos on day 1 of our desktop app huh even codex didn't do it

opencode has 2.5 (i'm the 0.5) people working on the core + tui we all use AI but nothing crazy, no skills, no plugins, etc we'll add one more person to takeover for me isn't it weird the companies claiming to have solved software need 10x than this

opencode's minimal mode will be radically simple we already have a full tui so don't need to get fancy - instead it'll feel like ai is just driving your shell this avoids the glitches you see in agents stuck in the awkward middle we started by refreshing opencode run

scariest commit message you'll ever see

all of our inference providers for kimi k2.5 are overloaded and asked us to scale down even after all this time there's still not enough GPUs

someone ai PR spammed moving off opentui to ink ....what???

why do most businesses stay so narrow once you have a machine going and a team in place why limit yourself why isn't spotify building an ai coding agent before you say focus remember how aws came to exist

we setup a beta channel for opencode you can opt in with `bun/npm/etc i -g opencode-ai@beta` right now we're testing our sqlite migration - if you run into any issues reply in this thread

"who cares if it's slop it works code doesn't matter" isn't a good way to sell ai im producing cleaner codebases than ever because of ai now i can actually refactor everything when we find a better way of doing something before shitty code just lingered forever

people get so mad when i'm harsh on vibe coding then they get really confused when they realize i work on opencode which makes them even madder

you can vibe code all of it who cares lie to yourself pretend it's a clever strategy but everyone will know they can feel it in your work how lazy you've become how little you care because how you do one thing is how you do everything

when we started opencode i spent a week trying every possible TUI library nothing met the bar and we realized we had to build our own do you work somewhere that could make this decision? or are you gluing together crap and pretending like it's a purposeful strategy

i'm pretty sure anthropic will one day make a model that's not available in the api and locked to claude code and i bet they'll say it's for "safety" reasons

it's worth having a bit of patience when a new model drops just in the last 24 hours can see how tps, ttfb, and errors have been drastically improving for kimi k2.5 in zen

who's really good at terminal UIs?

the thing i keep in mind is how crazy the crypto era was and how coinbase was the biggest winner and they did so by being normal while everyone else had blockchain psychosis or obtuse skepticism it's the same thing again

ai is such a magnifier of judgement that it's going to get really obvious who knows what they're doing and who doesn't

was testing invoking skills as a slash command and did it on a blank session llm saw it and was like "oh you obviously want me to take this info and put it into agents md where it belongs" ... yeah he right