Another day, another happy @brewdotnew customer 🥰

We launched Brew on Product Hunt this morning and we're #1 right now. If you've got 10 seconds, your support would mean a lot. ❤️‍🔥 producthunt.com/products/bre…

hey guys we're launching on Product Hunt today! if you want to make some cool on brand emails check us out and pls give us an upvote :) producthunt.com/products/bre…

Make people _feel_ something Shout-out to @Spotify for not succumbing to the pressure of mediocrity In this era, not taking risk is the biggest risk

Want to join a fast-growing start-up? Here are 3 early-stage startups you should check out: - @brewdotnew - @reer_ai - @pletor_ai Find out what they're building (and how to join) 👇

@brewdotnew is almost here. 👀 Think Lovable — but for Email Marketing. Here’s a sneak peek at Step 1: - Problem: Your marketing needs to be on-brand. Always. - Solution: Drop your website URL. In just 30 seconds, Brew extracts your entire brand design system — colors, fonts, tone of voice, and more — and builds a custom AI knowledge base for your brand. The full launch is just around the corner. Stay tuned. 🔥

youtu.be/izS97GXPtvE Learn how to add DNS records to @brewdotnew And oh... If you were wondering how we made the video: - @screenstudio for the recording - Nano Banana to create an image swapping CEO for Batman - Veo 3 to animate Batman - @elevenlabs for Batman's voice sound effects - @midjourney to animate the transitions WDYT?

Did you see us at @HubSpot's @INBOUND? 👀

llama 4 on groq: - pick best image from 20 options based on context - filter RAG results before feeding to larger model - summarize documents web pages into key points before creating embeddings for better retrieval - analyze user intent to build detailed prompt for larger model gpt-5: - generate email copy that converts - write chat clarifications responses claude: - build design edit marketing email code & automations

how we use llms in production: - llama 4 on groq → classification, filtering, routing, summarization (fast cheap) - gpt-5 → all user-facing text generation (cheaper than claude) - claude → anything involving code - gpt-image-1/flux → image gen (changes frequently) for coding: mostly claude and some gemini/gpt-5. smaller models for function-level changes only. general rule: use the cheapest model that works for the task except for user-facing text examples:

Quality of prompt is quality of thought

A pattern that works well for improving prompts: you: why did you do [X] incorrectly? llm: you: revise the prompt to prevent this from happening again

How we ship fast with enterprise quality at Brew with a lean team: 1) CEO comments on live app via Vercel toolbar 2) AI agent writes first draft opens PR 3) 10x engineers @cosjohh or @thomaspark_gg review & ship (usually <1 hour)

What if you could make code changes directly through the Vercel toolbar and have it spin up PRs for you?

Live crawling map. React flow AI SDK Microlink

My operator anti-portfolio: - Ramp at <20 people (had an offer, decided to join Nova Credit as COS instead) - Lovable at 4 people (had a bunch of convos with Anton, but decided to start @brewdotnew instead) - Bridge very early on (had convos with an external recruiter, but decided not to pursue) There’s no other option now than to make @brewdotnew a decacorn. Massive fucking chip on my shoulder