nk

building literally nothing, for now

[nk @nicongdev]

@steipete very helpful, marked as the first thing to do when i get 7 digits budget on tokens per month

[nk @nicongdev]

@hd_nvim @OpenAI @github lfg

[nk @nicongdev]

got these macro kb recently, honestly i wish they came with well designed and tested skills instead of making me set them ip myself

[nk @nicongdev]

# source mode markdown

> inspired by @obsdmd

https://t.co/UsBPYnEYtD

[nk @nicongdev]

gpt5.5 has been really bad lately at following my instructions, style preferences, even basic context i give it. something must have happened to it

[nk @nicongdev]

harness builders, why doesn't opencode just use UIMessage from the AI SDK, or at least extend it, since it's already so tightly coupled to it anyway

[nk @nicongdev]

just for the web app specifically, https://t.co/tBUu3Y6Iik is apparently more powerful than https://t.co/LMOYZy6GFm

i'd rather use free plan https://t.co/tBUu3Y6Iik than pro https://t.co/LMOYZy6GFm

[nk @nicongdev]

b50xX0nru

[nk @nicongdev]

Introducing LatentSys. See data through LLM's eyes.

everything's fake, just showcasing its SO easy to verify your startup idea nowadays

(latentsys was one of my failed ideas)

[nk @nicongdev]

@benln https://t.co/sLzx0StSXB

[Theo - t3.gg @theo]

My product is OSS and makes me $0.

I didn't mention it once during my Mythos video, because it wasn't relevant.

I did mention it in this video because it was relevant.

My "bias" is that I care a lot about good UX in agentic coding GUIs. I've cared about that since way before T3 Code.

T3 Code explicitly exists because the other options are not good enough. I feel this way about Codex App, Conductor, Cursor 3 (glass), Claude Desktop and every other thing I've tried.

Claude Desktop is still the worst one I've used, and it's genuinely insane to me that it shipped in this state 🙃

I don't fault you for it, I fault whatever system resulted in this being rushed out with too little eng time put into it. It's unacceptable that a $400b+ company put so little resources into this project.

[Nico Albanese @nicoalbanese10]

Deep dive Into how Open Agents can keep running for hours!

[nk @nicongdev]

i just had the early/rough version of nk-ai (my own agent harness) look at the harness AX (Agent Experience) around bash errors, and it pointed out that error messages are sometimes too long, which is bad AX. got me thinking about using a lightweight agent to summarize errors and keep context lean, but that adds latency

then i remembered using another product, which was just so slow like no matter what i wanted to do it took forever

some features, like summarizing a cli error message with a lightweight agent before passing it to the main agent, take longer and just don't work for real-time agents

but we don't always need real-time interactivity. sometimes we intentionally want the agent to run longer and produce better results

i used to model this as single completion vs workflow / background generation back in the early days of agents like ~2023. it's still a similar distinction, just a bit different from back then

let's call it interactive vs long-horizon

the clearest difference between the two is whether the harness should show the agent trajectory to the user

for interactive agents, show the trajectory but keep it from being too noisy. for long-horizon agents, always hide it

back to time-consuming features: use them as much as possible for long-horizon agents, and avoid them for interactive ones

[nk @nicongdev]

https://t.co/7Mmz0t16L9 was one of the projects that opened the door to modern TypeScript for me

it’s the codebase I default to whenever I want a reference for how to do something in a large codebase, and I think a lot of other devs did the same. it taught the community a lot

although I understand the reason and respect the decision, it still feels sad that they’re going closed source

[Bailey Pumfleet @pumfleet]

Open source is dead.

That’s not a statement we ever thought we’d make.

@calcom was built on open source. It shaped our product, our community, and our growth. But the world has changed faster than our principles could keep up.

AI has fundamentally altered the security landscape. What once required time, expertise, and intent can now be automated at scale. Code is no longer just read. It is scanned, mapped, and exploited. Near zero cost.

In that world, transparency becomes exposure. Especially at scale.

After a lot of deliberation, we’ve made the decision to close the core @calcom codebase.

This is not a rejection of what open source gave us. It’s a response to what risks AI is making possible.

We’re still supporting builders, releasing the core code under a new MIT-licensed open source project called cal. diy for hobbyists and tinkerers, but our priority now is simple:

Protecting our customers and community at all costs.

This may not be the most popular call.

But we believe many companies will come to the same conclusion.

My full explanation below ↓

[nk @nicongdev]

@pumfleet @calcom is commercial open source dead too?