Winston B.

🎯 Optimize your timeline into a precise prioritization system. Live & Free. In my previous post, I used a method to help me target which posts to reply to in order to maximize exposure. Knowing which posts had the most leverage, I could focus more on quality. The Logic: Math > Vibes. Tweet engagement follows a power law: explosive growth in the first 1-2 hours, then rapid decay. Most tweets are 90% spent within 3-4 hours. This extension calculates remaining runway by comparing current engagement density against the expected decay curve. Reply when there's still gas in the tank. NOTE: This is meant for indicating which posts are popular, not necessarily for the exact number of impressions estimated. The Tiers: 🚀 Breakout → 🔥 Hot → ⚡ Active → ◻️ Fading → ◼️ Spent Key feature: Filtering. It dims tweets based on what you want to filter out, so your timeline becomes a prioritized queue as you scroll. Reduce cognitive load. 100% local. No API. No data leaves your machine.

"Anywhere from $40 billion to $150 billion of leveraged loans packaged into US collateralized loan obligations could be disrupted by the AI boom," per JPMorgan

Cursor’s Composer 2 is likely built on Kimi K2.5. The model URL + tokenizer are strong signals. I love this direction: companies mid-train and post-train on top of OSS LLMs. Prediction: open-source model labs will monetize by taking a cut when others build on top of their models and scale to millions of real users. They will enforce this via licensing. That’s the flywheel. That’s how open-source AI thrives.

Yep, Composer 2 started from an open-source base! We will do full pretraining in the future. Only ~1/4 of the compute spent on the final model came from the base, the rest is from our training. This is why evals are very different. And yes, we are following the license through our inference partner terms.

Cursor’s new alpha product, Glass, shipped 9 months late and is a case study in the innovator’s dilemma. The inverse of what’s happening at Codex is exactly why I’m bullish on OpenAI. 9 months ago, I did a lot of user research on Claude Code as it started gaining traction. The signal was clear: people loved running agents in a separate terminal surface, but the lack of UI created friction. We built a new agent control plane, separate from the IDE, called Agent Window. It felt like the natural next interface to work with agents. Then we got a mandate from above to ship it as a part of the IDE and not as a separate window. That broke the model. Writing code and orchestrating agents are fundamentally different jobs. Developers still needed both, and collapsing them into one surface diluted both. What shipped instead was Agent Mode inside the IDE, a watered-down version of the original vision. By launch, the pitch was how similar it felt to the IDE, which missed the point entirely. Now, 9 months later, Cursor Glass is here. But the window has already shifted. I talk to dozens of companies every week, and most don’t even mention Cursor in their AI coding stack anymore. It’s Claude Code and Codex. Cursor is still widely used, but as an IDE, not a coding agent. Meanwhile at OpenAI, the Codex App started as a hackathon project. The team saw the future and just shipped it. Now it’s used by millions of developers. You can just build things. You should just ship things.