The second day started on top of a scar from the first: the automations built on the previous stream had a bug where leads were exiting the automation by hitting the goal without respecting the configured conditions, and he spent a good part of the day debugging that before the camera was even on. Still, the scoreboard already showed something concrete — R$337 in email marketing sales, the first proof the business works. The goal for the day was clear and specific: build the email marketing send engine inside Email Hacker AI, the flow that lets an email manager never have to open Active Campaign again.
The build happened on top of a swarm of agents running in parallel — five, six terminals open at once, each one owning a step of the wizard: one handling segmentation, another the product, another the knowledge base, another email generation, another the send itself. Since the official Active Campaign API doesn't expose endpoints for creating automations, he had to reverse-engineer the API itself to build a custom MCP capable of creating funnels, email sequences, and one-click automation upgrades — something, he says, not even the official docs teach.
Mid-stream, he stopped to explain the reasoning behind the whole architecture: the thesis that a vertical AI agent — not just a tool, but a replacement for the entire team that runs a traditional SaaS — grows much faster than conventional software, because instead of just selling the software it absorbs the tedious, repetitive work of a manager, copywriter, analyst, and dev. That's when the "About" page went up, explaining to the audience that Email Hacker AI doesn't compete with Active Campaign, it builds an autonomous-agent layer on top of it.
The tensest point of the day was technical: trying to feed the knowledge base (RAG) with YouTube video transcripts, the scraper hit a block, and he decided to buy a residential proxy live to get around it — paying on the spot, confirming the card, hunting for the API key in a panic. In parallel, the email generator kept getting things badly wrong: asked for an email about a kitchen product using a knowledge base about deliverability, and the result came out with zero connection between the two. The fix came with a copywriting rule — 80% content, 20% offer, always factoring in the avatar — which finally got the agent to connect the knowledge base content with the right offer.
The stream closed with the first real test send going out clean, and the final send scheduled to run for the whole base afterward. Along the way, the community grew too: a WhatsApp group was created and linked on the homepage, audience suggestions (like a VPS metrics panel in the super admin) went straight into the "brain" backlog, and the promised live push-ups happened when the audience hit the agreed number.