[email protected]

role-model is here: a model router for Pi that works with (almost) any cloud model as well as local models. Set up your pool of models, whether GPT 5.5 and DeepSeek V4, or Qwen 27B and Qwen 35B running locally, or combine and route between your local model for general chat and tool use and cloud models for more difficult tasks. role-model is a router runtime and and a forward-looking routing protocol that can enable consumer applications like Pi to classify its requests with required capabilities, task-type and suitable roles for better matching between task, role, and model. pi-role-model is a package for Pi that lets Pi manage and interact with role-model. role-model works perfectly with Pi even without the package - just configure your models in role-model, then configure role-model as your provider in Pi and choose routing alias. Install the package for Pi: ’pi install npm:@try-works/pi-role-model’ Then let your agent install and configure the role-model runtime or point it to: ‘github.com/try-works/role…’ role-model is in an early alpha release - let me know bugs, issues or usability questions. also, tell me if it just happens to work perfectly for you! there's lots more to do and you can help by installing and running role-model, and by reposting this tweet! thx

you ever heard someone say something was "lit" but you didnt know what it meant? i'll tell you, it means role-model github.com/try-works/role…

@DanielMiessler they also need to have a Slack competitor so they can put a bot in there and you can share google docs and other files in the chat etc

I don’t think most people realize how utterly strange it is that Google does not have an AI harness that is competing with OpenAI and Anthropic. Their ineptitude at product management has now gone from hobbling a company that was guaranteed to win to exposing it to existential risk. This is a company worth trillions of dollars that is medically unable to ship a product. Using any of their services as an administrator is the same type of torture that it was 15 years ago. Google’s inability to fix this should be studied in business books for decades to come. Starting now. It is the single strangest thing I’ve ever seen in business. They literally invented modern AI, and all they have to show for it is annoying pop-ups in Gmail and Google Docs that make everyone want to vibe code an alternative. The best evidence that ASI already exists is a theory that it’s at work inside of Google already, making sure they lose. What an absolute abomination.

@lucian__03 Just switch to 5.4 and it becomes unlimited

여러분!!! Codex 무제한 플랜, 진짜 있었어요!!! 방법: OpenAI 입사. 입사하면 스샷처럼 5시간/주간 사용량이 둘 다 무제한으로 뜬대요 ㅋㅋ 세상에서 제일 가입하기 어려운 플랜..

@GergelyOrosz Recursion. Recursive-mode.dev

What is "loop engineering" to you, anyway? Been looking into this, and also experimenting myself... and I'm not (yet?) buying that it's a new paradigm. But I might just be missing something. Can you share "loops" you regularly use?

@JinjingLiang will be very costly for everyone involved when Anthropic fails to IPO

The labs have exhausted all the PhDs... YC partners are apparently next. Taking leave from YC can't be cheap. The opportunity cost of leaving YC (even on leave) must be enormous. So this is quite interesting to watch. 👀

@kunchenguid Already built a stateful internal harness that's a lot less overengineered than that thing. Recursive-mode.dev

firstmate just crossed 1k stars on github! 🎉 it grew organically as a new way of working with agents. everyone i've talked to that started using firstmate told me they're getting a lot more done and enjoy the experience now let me tell you a funny story - up until today, i didn't even know what to call it! it's not a harness - you can run firstmate in almost any agent harness. it doesn't require you to change the tool you currently use it's also not a skill - it goes much beyond a skill in terms of shaping the overall agent behavior. and it's obviously also not a model, not a CLI, and not a MCP i call it an "agent distro" because i feel it's closest analogy is a linux distro it's a directory made of a system prompt, some built-in skills and bash scripts that shape your agent's behavior and allow it to act as your firstmate this setup allows the agent to be completely self-aware and self-evolvable. it can answer any question about its own behavior, and it can change every bit of itself according to your needs i've only just realized that it may be a category-defining solution, as there's nothing quite like it right now. if you have been using firstmate, what do you think? is "agent distro" a good way to describe it? github.com/kunchenguid/fi…

@failcatcat Great, now we know we can replace you with 5.6

卧槽我真的汗毛直立， 我和gpt5.6一起审5.5pro写的paper，我在飞机上看的非常仔细，然后写了一些comment，gpt5.6和我给的comment基本一模一样

I don't even have 300 followers but I post like I have 3 million

@edandersen PMs become literal demons

Junior developers get a massive boost (0 to something at all), top tier developers get a moderate boost, but its mid tier developers who are too expensive and the boost does not help them enough I guessed this two years ago and called it "Developer IQ" in a now delisted video

@schrockn OpenAI has been in the lead since October

I've seen enough: After a weekend of working with Sol and Fable it's pretty clear to me that OpenAI has taken the lead on coding models. Congrats to the team. You cooked. Looking forward to Anthropic's next move!

@0xsachi Check out my github and read my blog. You won't be able to stop thinking about me

Unfortunately interviews are not the best way to vet ppl Have ppl found alternative ways?

@bendee983 first you need to define failure

This is an interesting point. Failure -> recovery -> should be part of any advanced AI system. But we can also look at it from a different perspective. If human intelligence relies on this loop of trial and error, what happens when we delegate all that work to AI? There are a lot of discussions around AI doing the menial work and humans doing the high-level thinking and decision-making. But the problem is that to think, you have to "do." So without at least doing part of the menial work (e.g., coding, writing, etc.), what learning opporutnities are we losing?

@phuctm97 did the same this saturday, but with good old 5.4

A friend of mine matched with guy on Tinder. They scheduled a date on the 14th. The guy just canceled it so that he can spend more time with Fable 5 right after this announcement.

@AtlasShrug1 high hardware costs is good for O and A. if those costs drop, not only do they have no moat, but they also have a massive issue with their cost structure and worthless datacenters.

I am not understanding how this AI trade can play out well. I saw Gavin’s tweet about lower cost open source models, which conflicts with Altimeter’s assertion that the frontier guys will continue to gain more share. Either way, the casualties will be brutal. If the frontier models continue to dominate, OpenAI and Anthropic do well, but costs remain high and the whole Jevons Paradox thesis is delayed. That doesnt seem like a great outcome given where valuations stand today. On the other hand, if the low cost open source models take meaningful share, Jevons Paradox kicks in, but OpenAI and Anthropic business models may be at risk, they may struggle to consistently make money, and therefore their trillions of dollars in future purchase commitments that are the cornerstone of the buildout of “the ecosystem” may not be money good, creating potential liquidity problems and death spirals with all of the leverage now in the system. To me it appears that the odds of a major disruption of some sort is extraordinarily high one way or the other. And it may be the next iteration of companies that are either tiny and private now or that havent even been started yet that will figure out how to make money from whatever wreckage transpires. Am I crazy or is this plane unlikely to land smoothly?

@Dr_Gingerballs @AtlasShrug1 if hardware costs go down, O and A crash. their only moat, if any, is training capacity.

You got it. The only way for any of this to go well for model builders is for cost of hardware to drop 99%. In that case semis crash. Otherwise, model builders continue to hemorrhage money until they die and then it all crashes. It’s crash either way. Value of compute divided by cost of compute is trending to zero.

@MaxForAI 闲鱼有门票

外交部发言人宣布： 2026世界人工智能大会暨人工智能全球治理高级别会议将于7月17日至20日在上海举行。国家主席习近平将出席大会开幕式并发表主旨讲话。 难怪今年WAIC这么严格的安保….

I think the biggest issue with open source is that the issues and prs seem to tend to become the roadmap. It's easier to just merge stuff and launch new features, then sit down and build a product vision, deny prs, and build your own stuff, or communicate it to contributors. then you get bloated.

@beginbot looks like recursive-mode. recursive-mode.dev

I couldn't sleep last night, spent hours working on a perfect software factory and I think I finally cracked the god loop

@ibuildthecloud @Heather81031325 TUIs exist only because GUIs take a TON of time to build.

@Heather81031325 it doesn't make sense. IDE were invented for a reason, you can't say that TUIs or terminals are just better. They have a purpose, but the currently TUIs have totally gone the wrong route. I blame neovim, seriously.

I swear, I just hate the TUI coding agents soooooo much. They are so ridiculously obnoxious and not at all what I want.

@HistorianZhang @octal it's a Chinese company

@octal Nah KFC in China is probably one of the most successful examples of localization. This is an annoying China internet thing

Worst thing about China this trip: KFC seems not content with just a WeChat mini app, but tries to get you to install a dedicated kfc app, no other way to order. Maybe I am translating incorrectly (hard since images, so ads get mixed in). 2/2 failed attempts. A hate crime.