in love with @t3dotcodes from @theo / @jullerino / @maria_rcks new orch-v2, inline HTML , other tools and MCP hard refactoring my OWN skill stack to be PURELY t3code based now github.com/hani-q/qstack/tre… This is how i actually decide on my Agent lineup.. Football / Soccer ( Ai / Si 😛) nerds will recognize the FM love i had growing up more cool stuff/integration coming soon

Theo never told me to hide anything hahaha. if anything he told me to be 100% transparent and that it didn't matter if t3 code ended up low i didn't explain it well, that's on me. cloudroom is still early access and putting it next to tools that have been on the market way longer didn't feel fair and cloudroom will be in the benchmark too, same tests as everyone, so the data decides. happy to show the chat if anyone wants it

Imagine getting mad that a shill yapper YouTuber shit out the best AI tool I’ve ever used.

I know T3 Code is good because there are now companies I’ve never heard of spreading lies about us to try and get users Literally every single thing he said here is a lie lol

I did not tell him to hide anything. I hit him up to ask about why T3 Code scored low. Learned he was on a 3+ month old version. He tried the modern version with our new orchestrator, realized his scores were not good, and adjusted. He mentioned wanting to make a real benchmark after, so I offered to fund it because there’s no good research like this going on right now. He actually found a few things I’ve been rabbit hole-ing on all weekend. I have no idea what Cloudroom is. I definitely did not tell him to remove it or anything else. Only commentary I had on other solutions was that I was curious why a shitty clone of our app scored higher than us initially. Nothing to do with cloudroom.

It is becoming f*cking dangerous: you can now reverse-engineer any repo with one command - a YC-backed tool with 127k stars maps the whole codebase, and your agent walks the map instead of reading files 1. Clone the project you want to understand 2. Install graphify: uv tool install graphifyy 3. Run graphify install (it detects Claude Code, Cursor, Codex or Gemini CLI) 4. Open the project and type /graphify . 5. Open graph.html: the architecture map nobody ever documented 6. Read GRAPH_REPORT.md: what everything flows through and the connections you didn't know were there 7. Ask: graphify path "A" "B" traces how two things connect, graphify explain shows everything one of them touches Code is mapped locally with no LLM, and every link says whether it was read or inferred Turn your agent into a reverse engineer that knows the whole project before it touches a single file Find the repo in the reply below

To clarify: Replit remains model agnostic. Our users can choose Anthropic models as well as many other supported models. What I meant at TC Disrupt is that OpenAI is now our preferred partner powering Agent based on performance, cost, and strength of our partnership. We believe in giving our users model choice while optimizing the default experience for the best performance at the best price.

This is how Sharia law begins… first groceries become cheaper. Then, housing becomes affordable. Then America falls.

Can I be honest for a sec? Regarding LLMs in gaming, I feel more confused than ever. And I would like my peers and other people who are familiar in this field to chime in. It’s barely been a week since I made my post about distancing myself from a vibe coded project made by my friend in Smash. I even said “AI has no place in gaming”. After I made that post, I was reached out to by multiple colleagues in the space and people hired full time in software development. They educated me severely on the topic and I feel dumb now for my earlier sentiments. I was informed by my colleagues that it is now the norm to do 90% (and this is low estimate apparently) of all your coding with AI assistance or vibe coding. 75% of all new code at Google is made by AI and engineer approved. The vast majority of all the apps we use now are updated or maintained with AI coding - I imagine that also goes for Twitter, the very app you’re reading this on. So I figured okay - it’s here to stay in all software development - but where does that leave video games? I saw the posts from Epic Games, XBOX, and other publishers that seem to embrace the new era instead of rejecting it. In the last week I’ve been sent about a dozen different projects like this one that dissect and reintegrate Melee in multitudes of creative ways. One of them is even a build of Melee that runs natively at 120FPS and one of them simulates the feeling of a CRT almost perfectly. Some of them feel silly and useless, like low effort projects someone probably gave a single prompt and an ISO of Melee to build. Actual slop. And yet some feel like legitimately useful things that the community or people practicing could benefit from. I’m doing my best to navigate what is clearly a rapidly changing sector, especially when we regard the question of “Are Video Games Art?” And since they clearly are art, does that render AI in its development a bastardized amalgamation of what’s been “learned” from people with actual original ideas? Or is it strictly optimal to use AI in game development because games ARE software and therefore a necessary tool to wield in this day and age? I am making this post because I need more guidance on this topic and I feel open discussion on it might be the best way to do that.

Two things can be true at once, and I think the argument is because of people confusing the two. 1. AI has no creative taste and makes slop when asked 2. AI can do incredibly tedious work to make awesome things possible For a really simple example: having AI write a novel would be terrible. AI translating a novel to a different language with human oversight is much less terrible. Wish people would stop confusing the two

I tried t3 code and it was what I was looking for. I didn’t know what t3 code is so I wasted handful amount of tokens to build it from scratch…

And t3 code is the GOAT among the agent control planes. Nice work, @theo!

OK, so: an elastic vol w/ fseventsd turned off. had to be elastic as I didn't want to pre-reserve space. TL;DR: created an APFS volume with nolog + disable Spotlight indexing at ~/repos - persists and looks like a normal folder to every app. script: github.com/elithrar/dotfiles…

I did not expect to feel this difference as much as I do

OLTP is expensive. You can ClickHouse it. To change ~100 bytes of a row, Postgres modifies several 8 KB heap and index pages and writes tens of KB to disk. That's the price of ACID, and you keep paying it for data that stopped changing months ago. Keep writes and fresh reads in Postgres. Serve settled data from ClickHouse, with 10–30x compression and aggregations at about 100x lower cost: clickhou.se/4yQe2bz

Sorry bro did you say something ?

Today, we are launching OpenRouter Armor in beta. Tool calls allow agents to interact with the world, and regex-based filtering often isn't enough to detect unwanted tool calls. Armor uses @typesafeai's Jev model to guard against out of policy tool calls before your agent executes them, with better performance than Claude Code's auto classifier at a fraction of the cost.

Introducing Inbox Zero Mail. The open-source Superhuman alternative. - Fast and keyboard-first - Unified inbox across Gmail and Outlook - Split inboxes - AI categorization and drafts - Bulk unsubscribe - Team collaboration Runs on web, desktop, and native iOS. Free to use. AI is $20/mo, half Superhuman’s price. Open source, so you can self-host for free too. This isn't an overnight vibe-coded project. I've been running Inbox Zero for 3 years: 12k GitHub stars, 25k+ sign-ups, and enterprise customers. Until today, Inbox Zero was an AI assistant for Gmail and Outlook. It still is. But now we have our own email app too. We're SOC 2 compliant, and have undergone multiple pentests, including recent ones from OpenAI and Snyk using unreleased cyber models. Try it: getinboxzero.com

i built something for this! github.com/rayhanadev/mds-by…

Today we’re releasing Muse Spark Open, the most capable US open-weight model to date. Built for coding and agentic work, it excels at long-horizon tasks, complex software engineering, and multi-step reasoning. Key Model Specs: → 816B total / 42B active parameters → 1M token context window → Natively multimodal across text, image, video, audio