One Jev pattern I’ve been playing with that I haven’t seen discussed much: each call deliberately leaves behind a small piece of state only for the next Jev call.
I’ve started thinking of it as Jev carry state.
I pass Jev the normal application state it needs to decide. Separately, I distill a few selected decisions from the previous call into a much smaller carry state that exists only as context for the next one.
I’m using an animated character as one testbed for this. A Jev decision might affect the actual animation, while the carry state preserves the semantic trajectory behind it: we’re settling from excitement, continuing this behavior, keeping the expression restrained, avoiding another large change.
Some of that never needs to become application state at all. It only exists to help the next Jev call understand the intended trajectory of the previous decisions.
Keeping that state small and curated preserves the trajectory without replaying the full history or passing every implementation detail back through the model.
The result has been surprisingly useful for temporal coherence. A sequence can continue, settle, change direction, or escalate in a way that feels connected to what came before.
Jev itself is still stateless. The continuity lives in the carry state that one call leaves for the next.
I suspect this pattern is useful well beyond animation.
Curious whether others are doing something similar. @typesafeai @CompleteSkeptic
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
Resolved, same day. Turned out to be a subscription edge case on my account colliding with the new Pro signup freeze. Not a ban, not a flag. Thanks to the @OpenAI team for moving so fast.
For everyone who said the numbers weren't real: ~$80k on OpenAI credits in the last 7 days, peaking at $27k/day. Another ~$80k on Anthropic since Sep 1. One person, one personal account per lab, one MacBook and no servers, every credit paid out of my own pocket, no company card, at the same rates anyone can get. Receipts below.
This is what building with agents looks like in 2026. What all this is building: soon.
I've been spending ~$20k/day out of my own pocket on GPT-6 Astra in Codex for the past week. One person, one personal Pro account, paying for every token, running long lived agentic dev loops around the clock.
This morning my one and only account was downgraded to Free with no email, no notice. Codex just stopped mid-task: `gpt-6-astra is not supported when using Codex with a ChatGPT account.` Billing is clean, not a renewal issue.
The new Pro account freeze means I can't even re-subscribe to get Astra back. @thsottiaux if an automated check flagged this I'll verify whatever you need. Just need a human on it. DM open 🙏
I've been spending ~$20k/day out of my own pocket on GPT-6 Astra in Codex for the past week. One person, one personal Pro account, paying for every token, running long lived agentic dev loops around the clock.
This morning my one and only account was downgraded to Free with no email, no notice. Codex just stopped mid-task: `gpt-6-astra is not supported when using Codex with a ChatGPT account.` Billing is clean, not a renewal issue.
The new Pro account freeze means I can't even re-subscribe to get Astra back. @thsottiaux if an automated check flagged this I'll verify whatever you need. Just need a human on it. DM open 🙏
Boy, have we got news for you! 🧑🍳
The @openwork_ai team is happy to announce:
- Amazon Bedrock integration, contributed by our friends at @awscloud (thanks guys!)
- Native @deepseek_ai integration
- Integration with @openrouter and @LiteLLM!
Here's how what it looks like >>
Today we are launching @openwork_ai, an open-source (MIT-licensed) computer-use agent that’s fast, cheap, and more secure.
@openwork_ai is the result of a short two-day hackathon our team decided to hack, which brings together some of our favorite open source AI modules into one powerful agent, to allow you to:
1. Bring your own model/API key (any provider and model supported by @opencode is supported by Openwork)
2. ~4x faster than Claude for Chrome/Cowork, and much more token-efficient, powered by dev-browser by @sawyerhood (legend)
3. More secure - contrary to Claude for Chrom/Cowork, does not leverage the main browser instance where you are logged into all services already. You login only to the services you need. This significantly reduces the risk of data loss in case of prompt injections, to which computer-use agents are highly exposed.
4. Free and 100% open-source!
You can download the DMG (macOS only for now) or fork the github repo via the link in bio (@openwork_ai).
Let us know what you think (or better, send a pull request)!
Tal Saiag retweeted
היום ב-Tel Aviv CTO Club יצא לנו לשמוע מ-@TalSaiag על המסע של @GranulateLtd, אחד האקזיטים המרשימים שהיו פה בשנים האחרונות ובכלל (הרכישה בידי אינטל ב-650 מ׳ - ע״פ העיתונות - בתוך פחות מ3 שנים).
היו במסע הזה המון אירועים מעניינים, אבל דווקא ההתחלה שלו איכשהו מפוצצת לי את המוח >>
Tal Saiag retweeted
The last one is just annoying 😩
Our R&D team shared frustrating scenarios when it comes to #code and #performance.
Have anything to add? Get it off your chest and share with us in the comments. The top ones will receive an illustration!
#production #developers #engineers #CPU
Tal Saiag retweeted
Learn how @GranulateLtd leveraged its validated #AWS skills to drive 300% YoY revenue growth & help customers cut cloud costs by 60% - go.aws/3A2EPno
Tal Saiag retweeted
We're are proud to announce Granulate's $30M Series B investment led by @RedDotCapitalPartners with @insightpartners , @dawncapital, @TLV_Partners, @HetzVentures - bit.ly/36BwSIo.
See Open Position - bit.ly/36G3dOx
















