sergio.dev
LinkedIn Get in Touch
Facet · Marketing

Agent Marketing

Marketing that agents run and numbers prove.

Agent marketing isn't generating a pretty post: it's closing the loop. The creative produces itself inside the brand's voice, the ad lands in a WhatsApp conversation, the conversation becomes an order, the order becomes a metric — and the metric decides the next creative. I build that entire machine and run it for a real business, my family's: when an ad misses, no slide gets hurt — the day's snack production does. A funnel instrumented end to end, cost per conversation around R$2, zero guesswork.

Growth Loops Click-to-WhatsApp BrandContext Agent-Made Creative Meta Ads Measured Funnel R$ per Conversation Lead Capture Session Replay Ayrshare API Growth Loops Click-to-WhatsApp BrandContext Agent-Made Creative Meta Ads Measured Funnel R$ per Conversation Lead Capture Session Replay Ayrshare API
Talk to me 📄 CV — Agentic Solutions Architect (PDF)
R$ 2.28 Cost per conversation started (CTWA)
0 Conversations generated in the World Cup campaign
0 Agent-operated creation and publishing tools
0 Adminer Analytics extension downloads

The laboratory is Salgadelli, my family's snack business. The loop runs whole: the agent creates content inside a BrandContext with the brand's canonical facts, voice, and constraints; paid media distributes it; the click opens a conversation directly with the service agent that knows the menu and prices; the order dispatches to the kitchen; and the in-house ads and conversion dashboard turns everything into data that picks the next round of creative.

The proof in numbers was the Click-to-WhatsApp campaign on Meta during the World Cup: R$1,134 lifetime spend, 498 conversations started, R$2.28 per conversation. No landing page in between — the click lands in the conversation that converts, the least possible friction between the ad and the order. And the creatives were published via API, with no manual studio in the way.

I started down this road in low-code: n8n and Make automating leads, content, and the family bakery's Instagram/ads — that's where I earned the judgment of when no-code is enough and when to graduate to production code. And I practice what I build: this site is agent-marketing dogfood, with conversational lead capture, homegrown session replay, and score-based anti-spam. The medium is the message.

Architecture

The closed loop: the brand feeds the agent, the agent creates and publishes, the click becomes a conversation, the conversation becomes an order — and the metric decides the next round of creative.

Projects

Through this lens

Growth · Closed Loop Live

Salgadelli Growth Loop

Growth loop: content → ads → WhatsApp → order → metrics → repeat.

The mother loop: growth for a brick-and-mortar business treated as an engineering pipeline. The agent creates, paid media distributes, WhatsApp converts, the order becomes data — and the data picks the next creative. No monthly PDF report: an in-house dashboard, round by round, with the owner seeing what every real spent turned into.

Impact
  • Full loop: creation → distribution → conversation → order → metric
  • Ads and conversion dashboard made to measure, not generic SaaS
  • The metric feeds the next creative — the loop learns
Stack
Ruby on Rails 8 Meta Ads WhatsApp In-house dashboards
Ads · Click-to-WhatsApp Live

Meta CTWA Campaign

Click-to-WhatsApp ads landing straight in a service agent — a fully measured funnel.

The shortest funnel there is: the ad click opens a WhatsApp conversation directly with the agent that knows the menu, prices, and promotions. A World Cup campaign measured end to end — R$1,134 lifetime, 498 conversations, R$2.28 per conversation — with creatives going live via API, no manual studio.

Impact
  • R$2.28 per conversation started · 498 conversations · R$1,134 lifetime
  • Zero landing page: the click lands straight in the conversation that converts
  • Creatives published via API (Ayrshare) — publishing with no human touch
Stack
Meta Ads (CTWA) WhatsApp Ayrshare (API) Rails dashboards
Creation · Publishing Production · Private

Creative Studio

Agent-operated creation and publishing studio: from brief to live post.

The machine's creative factory: 6 tools that generate the asset (Replicate), write copy in the brand's voice, and publish across networks via Ayrshare — all anchored in a BrandContext with canonical facts, voice, and constraints. The agent doesn't invent the brand: it obeys it. From brief to live post without passing through an editor.

Impact
  • BrandContext enforced: canonical facts, voice, and constraints per brand
  • Asset + copy + multi-network publishing in tool calls
  • From brief to published post with no human editor in between
Stack
Ayrshare (multi-network) Replicate (image generation) Claude Ruby on Rails 8
Low-Code Era Graduated to code

n8n / Make Automation

Before custom code: agentic automations in n8n and Make for real businesses.

Before the in-house harness came low-code: extensive n8n and Make workflows doing lead routing, content generation, data sync, and the Instagram/ads automation for the family bakery. It was the stage that gave me the judgment I still use with clients today: where no-code solves it, where it breaks, and when to graduate to code.

Impact
  • Marketing automation running for real businesses, not a sandbox
  • The family bakery's Instagram and ads on autopilot
  • Lived judgment on graduating: no-code → production code
Stack
n8n Make OpenAI Webhooks / APIs
Meta · Dogfood Live — you are inside it

This Site

The portfolio as an artifact: conversational lead capture, in-house session replay, and anti-spam.

I practice in my own funnel what I build for others: this site captures leads in a multi-step conversational flow, records homegrown session replay (zero tracking SaaS), scores spam with a honeypot + signed token + language heuristic, and serves the right CV to each profile. You are inside the case study right now.

Impact
  • Conversational lead capture with per-visitor state
  • In-house session replay and analytics — no third-party pixels
  • Score-based anti-spam and interest-segmented CVs
Stack
Ruby on Rails 8 Hotwire / Stimulus In-house session replay Heuristic anti-spam
Experience

The track record, re-read

Sep 2021 – Nov 2022

Software Engineer

Adminer (Remote, B2C)

  • E-commerce analytics at millions of requests/day — the raw material of every growth loop is reliable data
  • 'Adminer Analytics' Chrome extension with 50,000+ downloads: product distribution in practice
Node.js Ruby Angular PostgreSQL MongoDB Elasticsearch
Skills

The toolkit for this lens

🤖

AI Agent Systems

Multi-agent orchestration in production. Tool-call loops, sub-agents, autonomous Claude Code workers, MCP servers, context engineering.

Multi-Agent Coordination MCP (Model Context Protocol) Function Calling / Tool Use Claude Code SDK Sub-Agents & Hand-offs Context Window Management
📊

Data & Vector Stores

Schema design, query tuning, RAG over proprietary data. pgvector-first, lifted from production scars.

PostgreSQL (tuning · partitioning) pgvector / RAG BigQuery · Looker Elasticsearch MongoDB · MySQL ETL & Data Migration

Want a growth loop running on its own in your business?

End-to-end agent marketing: creation, media, conversation, and metrics — built, instrumented, and proven on a real business.

Talk to me 📄 CV — Agentic Solutions Architect (PDF)
SB
Sérgio Brito
Online
👤 About 📚 Projects 💻 GitHub ✉ Get in Touch