I rebuilt "Build Your Own AI Assistant" from scratch as a structured 13-step tutorial — from a bare Gemini API call to a fully deployed coding agent with Telegram, persistent memory, and sandboxed cloud execution.
Based on the work by @ivanleomk and @hugobowne.
🧵👇
What I added on top of the original:
- Unified step-by-step progression
- Dynamic tool rewriting that works in cloud
- Sandbox persistence actually connected
- Deploy/teardown scripts for every step from 09+
- A presentation walkthrough
Next step for me: level up my software engineering + system design so I can pair with agents even better. If you’re thinking about the next cohort, here’s the link: https://t.co/5CsFzyNwV9
Big thanks to @vig_xyz and @jxnlco for the AI Coding Accelerator on Maven. I joined thinking I’d get a few tips—walked away with a totally different mental model for building with AI coding agents.
Balancing autonomy with direction is the cheat code: let the agent move fast, but keep architecture and final decisions human. It’s boosted my learning speed—and forced me to explain what I want with way more clarity.
@leerob How to dynamically inject information about current date using universal way like https://t.co/5nF8WbHj8R or something else (not cursor rules), because often web search uses year 2024 even if you specify date in chat message?
🌟 Announcing the LangChain Experts Program — and our founding cohort!
Big shout-out to @labdmitriy , @mchonedev, @PawelTwardziak and Mohammad Mohtashim for leading the way.
Experts help users succeed by answering questions in the Forum, and contributing across Slack + LangChain Academy.
Meet the cohort → https://t.co/WQaY13YXCA
📷
It’s sad to see how chasing revenue seems more important than respecting customers. Whether they cancel the subscription or not, once my current period ends I will never use @higgsfield_ai again.
Sharing my experience with @higgsfield_ai. On Nov 21 I bought the Ultimate annual plan because the site said it included 365 days of Nano Banana Pro 4K. Right after I paid, the offer changed to only being valid until Nov 28, 2025 and 4K vanished from the Ultimate plan.
What makes this worse is that @higgsfield_ai support replies read like generic LLM templates: polite and “caring” but not actually resolving anything. They used to answer in ~5 minutes, but after I mentioned my suspicion about LLM-generated replies, they stopped responding.