Vinicius Flores, PhD

Work / AI & operations

Marketing operations with AI agents

Period
2025–present · Bibliotheca and this site
Design
Workflow design, briefs, and the source material each agent is allowed to use
Control
Review standards and the final check on every output

Challenge

A small central marketing team, a website in 14 languages, four analytics tools whose numbers lived in separate spreadsheets, and a growing list of content and projects.

None of the work was difficult to define. There was simply more of it than anyone could do by hand.

Approach

I treat AI agents the way I would treat a new colleague: they get a brief, the source material they are allowed to use, and a reviewer. Every workflow produces something a person can check line by line, and I do that check before anything is published or acted on.

How an AI agent workflow runs, with a human check before anything shipsFive steps: brief, approved sources, agent, human review, publish or act. Human review can send work back to the agent for revision.01Brieftask, audience,standards02Approved sourcesmanuals, glossary,data exports03Agentdrafts, audits,analysis04Human reviewline-by-linecheck by me05Publish or actsite, Asana,reportsrevise and resubmitNothing is published or acted on before step 04.
How an AI agent workflow runs, with a human check before anything shipsFive steps: brief, approved sources, agent, human review, publish or act. Human review can send work back to the agent for revision.01Brieftask, audience, standards02Approved sourcesmanuals, glossary, data exports03Agentdrafts, audits, analysis04Human reviewline-by-line check by me05Publish or actsite, Asana, reportsreviseNothing is published or acted onbefore step 04.
How each workflow runs. Agents work from a brief and approved sources; a person signs off.
Analysis

Four platforms brought together to inform a growth plan and assess website performance.

Content

Messaging reviewed with synthetic audience panels, claims checked against the manuals, and copy structured for AEO.

Operations

Localization audits, project setup, and daily planning run as agent workflows.

Work completed as part of my role at Bibliotheca and on this site. Commercial data withheld.

Connecting marketing data to pipeline

Google Analytics, Search Console, Google Ads, and HubSpot each tell a different part of the story, and none of them tells it alone. I brought detailed exports from all four into a single analysis and used agents to work through years of data and follow it to what matters: which activity produces sales pipeline.

That analysis is now the basis of a growth plan and a full review of our digital marketing, and it is where the before-and-after study behind the website case came from.

Writing for people and answer engines

For the new website, I structure product copy around buyer questions and objections, with clear explanations and FAQs as part of the answer engine optimization (AEO) work.

I use synthetic audience panels to explore how different buyers might respond to the messaging, then check product claims against the technical manuals before publication.

The same workflows cover localization: the 14 language versions of the site are audited against the company glossary to find terminology used inconsistently from one market to the next.

Planning and delivery

Planning and project setup run as agent workflows: the editorial calendar became a structured project in Asana, and I use Claude with Asana to organize my own priorities, which keeps tracking and admin from taking over the week.

This site was built the same way, through an agent connected to WordPress, from structure to publishing.

An agent can draft, compare, and check faster than any of us. Deciding what is true for our customers, and what is worth publishing, stays with me.

Back to all work