Most trading ideas never make it to a proper backtest to Deployment cycle . We have curated a solution to change that - Agent Z Describe your strategy in plain English, and Agent Z helps turn it into a structured workflow across research, backtesting, validation, and deployment readiness using AI agents and frontier models. As a closed beta user, you get early access to an AI-native trading workbench designed to help you test strategy ideas in a fraction of the time. You bring your market knowledge, Agent Z brings the agentic workflow, model access, coding support, and connected context. The best part - you stay in loop at every step. Apply for early beta access: tally.so/r/OD8ANK

3 failed launches. Almost bankrupt. Still kept going. Then SpaceX became the first private company to reach orbit and changed the future of space. Some stories are not about luck. They are about belief, obsession, and insane execution. Congratulations Team! #SpaceX

Food for thought!

Fable or 5.5 If you planning to use loops … none ! Unless you have access to infinite token managing workflow is a nightmare…

Agent x Loop It should have been obvious 3 years ago to that putting an LLm in a loop was the endgame. I’m surprised y’all weren’t already thinking about ai coding like this Your job is not to guide a coding agent It’s to surf on the back of a bucking bronco demon alien that’s smarter than you and can work in 1000x parallel That’s an entirely different task from writing clear linear tickets or product descriptions for small features You don’t tell an agent to pick up a hammer and assemble a chair with a detailed design spec. It already knows to do that. Your new job is building onboarding documents, design direction & high level cultural values (aka decision frameworks) for a furniture factory, distribution center & business Then you simply hire Ai workers on loop to do the thing It’s very similar to the way fast food restaurants have standardized and documented practices to teach dumb unmotivated and non-sober high schoolers how to run a fully working restaurant, at the scale of millions Except instead of producing burgers or chairs you produce code & outcomes via software

Wild how quickly AI usage compounds. 2 weeks after Goals launched, my daily token consumption is already up 3x. Better compaction has made long-running chats feel natural. 100M token conversations are becoming normal.

One good practice is to set up and maintain a validation chat (a session where it acts like team representation and you can discuss and understand what happened) along with a loop. Since the memory and base repo for both the chat and loop remain the same. You can use /slide while it’s working and skills like /grill-me-docs or /teach. The user gets a clear picture of what’s done and how we think for next steps! I have seen loops descend into never-ending rabbit holes, but the good thing is the latest models save us from creating completely useless slop. They give something relatively relevant and mostly a too-specific fix.

this is good

I want some kind of LLM workflow tool. • Ability to manage a set of input files (Markdown or similar), plus other general-purpose context. • With real-time collaboration. (And maybe some concept of snapshots or VCS integration.) • And the ability to create/manage a inference workflows and a stored set of prompts. • Access to general-purpose coding agents (and not just chat models). • Some concept of compiled outputs/inference results (which ideally can be shared externally). Many projects have this feeling: "there is all this stuff, which I want to process/compute over in this iterated way, with some build artifacts being important/worth saving." GNU Autotools x Notion or something. Is anyone building this?

Saying that LLMs are just next token predictors is underselling these beasts to a mind numbing degree. First, LLMs aren’t just predicting the next token. They plan ahead - the loss function is the average of cross entropy across all future tokens in a context window and the attention has access to all previous tokens. So, at a particular token, the LLM is planning what could be relevant far ahead and not just at the immediate next token. Second, LLMs are trained to predict sequences across all texts on the internet that contain not just human generated text but things like weather forecasts, financial series, code, bash dumps, satellite pings and so on. To be able to do this prediction well, LLMs have to infer the physics / dynamics for all such domains (e.g. to predict weather patterns in data, you need to develop a model of earth coordinates, sunlight patterns, monsoon cycles and so on). (Just think how hard this prediction problem is, given the diversity of texts in the pretraining corpus.) So, instead of saying LLMs are next token predictors, a much better way of framing them is that these things are universal simulators. You prompt them with a hint, and they simulate physics of whatever domain your hint is about It’s crazy scary that this works but somehow it does!

you could build a top tier venture firm just focusing investment decisions short and long term based on deep model benchmarking / evals find capability overhang, find areas models suck and track trajectory, etc

Token costs are why there will be no saas apocalypse / good dev tools are cached intelligence for agents! The popular theory goes: agents can write code, so they'll just rebuild every tool from scratch and hit raw APIs. no more dev tools, no more CLIs, no more software layers. just agents and endpoints! We just tested this and the data says the opposite. We benchmarked Claude Code and Codex on real Hugging Face Hub tasks (~1,000 graded runs), with two setups: the agent-optimized hf CLI vs the agent hand-rolling curl or SDK calls from scratch. Hand-rolling burns up to 6x more tokens on multi-step tasks and fails more often (84% vs 94% task success). And that's just dropping one abstraction layer. It would obviously be orders of magnitude more tokens and a dramatically higher failure rate if the agent tried to bypass HF altogether and rebuild model hosting, versioning, and distribution from scratch. Every time an agent re-derives a workflow from raw API calls, you pay for that reasoning in tokens. every single run. a good CLI compresses that entire chain into a few high-level commands the agent can't get wrong. In a world where everyone is complaining tokens are too expensive, abstraction is leverage: thousands of hours of design decisions your agent doesn't have to re-reason about at inference time. Good tools are cached intelligence for agents! So no, agents won't rebuild everything from scratch. they'll gravitate to the most token-efficient tools, because that's what their owners pay for. The software that survives won't just be accessible to agents, it will be accurate and cheap for them to drive. We're seeing it happen with HF, which is becoming the platform for agents to use AI: ~49M requests in just two months, and growing fast! huggingface.co/blog/hf-cli-f…

Hot take: Universities charge $300,000 for a degree that teaches you skills any LLM can do for free. At some point we need to have an honest conversation about whether higher education is the greatest individuals misallocation of capital in recent history.

PRO TIP: Gamify your notebooks Don't just read your notes— investigate them. Our new Sherlock Holmes notebook turns studying into an interactive mystery game. Deduce facts, uncover clues, & prove that even the most complex matters can be elementary. ➡️ goo.gle/Sherlock

We’ve been researching new ways for ChatGPT memory to carry context across conversations and keep it useful over time. Today, that work is rolling out as a more capable memory system in ChatGPT. openai.com/index/chatgpt-mem…

Wrote something small today. Some lessons don’t come with a syllabus. You only understand them after you enter the field and let reality teach you. open.substack.com/pub/neelsi…

It's time to fly.

Dreams don’t need permission. They need belief, discipline, and action.

I think codex The best openclaw set up with UI of the most reliable quality! With the models edition and a few iterations. I will definitely close down my claws and use codex to get things done end to end. The app is getting better daily It just needs to create less .codex storage files