Federico Jan

Enterprise AI Agent Architect · GTM Engineer

Federico Jan

I design and deploy AI agent systems for real business operations.

Most recently, I was the first engineer and sole agent architect behind REMI, Marketing.MBA's AI GTM operating system. I took agent systems from architecture to production and worked directly with employees as they met the realities of the business.

Explore the flagship systemStart a conversation →
Portrait of Federico Jan
Federico JanAgent Architect · GTM Engineer
Tokens in production
10B+
Frameworks in production
5+
Agents developed
20+
Agent runs traced
Thousands

Selected work

One flagship platform. Two focused systems.

All work →
  1. Flagship systemBuilding an Enterprise Multiagent GTM Operating System

    The agent platform behind production GTM work: an internal agent factory, structured company context, specialist teams built for long runs, versioned skills, observable execution, approval gates, and business integrations.

    • 10B+ tokens
    • 5+ frameworks
    • 1h+ runs
    Read case study
  2. Building an Agent Native SEO Department

    A manager agent and three specialists turned Airtable briefs into Framer articles approved by humans and ready for production, including research, detailed copy, SVG infographics, video embeds, and internal links.

    • Agent native SEO
    • 4 agent system
    • ~90 min/article
    Read case study
  3. Ask Your Org: A GitHub Native Company Brain

    A governed organizational knowledge system that turned scattered company context into a navigable GitHub structure with explicit sources, ownership, freshness, and escalation rules.

    • GitHub native
    • Provenance aware
    • Org intelligence
    Read case study

Public blueprint

Make your organization legible to AI agents.

The public Ask Your Org Blueprint turns company context, source ownership, permissions, and escalation into a practical three layer architecture.

Get the blueprint Explore the proof

Operating standard

Enterprise grade means surviving the business.

The hard part starts after the demo: observability, reliability over long runs, systems integration, and adoption by the people doing the work.

Production Agent Operating LoopSense → decide → act → measure
Production Agent Operating LoopMarket and customer signals move through manager and specialist agents, cross a human approval gate, become operational action, and return as traced outcomes for the next run.PRODUCTIONLOOPTRACED, NOT ASSUMEDSenseMarket & customer signalsDecideManager and specialist agentsActHuman approved executionMeasureTraced outcomesProduction Agent Operating LoopA compact mobile view of the sense, decide, act, and measure production loop.PRODUCTIONLOOPTRACED, NOT ASSUMEDSenseMarket signalsDecideAgent teamsActHuman approvedMeasureTraced outcomes
Signals enter the system, agents coordinate, humans retain authority at consequential edges, and traced outcomes improve the next decision.
  1. 01

    Instrumented

    Thousands of runs traced from the beginning, so failures became engineering evidence.

  2. 02

    Long running

    Agent workflows designed to stay coherent and stable for more than an hour.

  3. 03

    Integrated

    CLIs, APIs, knowledge systems, approval gates, and production GTM workflows.

  4. 04

    Forward deployed

    Built alongside employees and tested against real incentives, friction, and constraints.

See the operating system behind the work →

Operator perspectives

What production taught me.

Technical library at Starkslab →
  1. Five Decisions That Made My AI Publishing Pipeline Trustworthy

    Why I chose local markdown, an explicit approval gate, separate prompt files, durable fallbacks, and verbatim evidence for my publishing workflow.

    Read perspective
  2. My Standard for Production Ready AI Agents

    What tracing thousands of runs and deploying agents inside a real business taught me about control, observability, approval, and recovery.

    Read perspective
  3. Why I Build Agent Factories, Not Fake Teams

    After building more than 20 agents, I learned that reliable multiagent systems behave less like artificial teams and more like governed production lines.

    Read perspective
  4. What 13 Autonomous CLI Calls Taught Me About Agent Tooling

    One unscripted agent run exposed two bugs and changed how I design command line tools, structured output, errors, and recoverable workflows.

    Read perspective

Building or hiring for production AI?

Open to AI leadership roles and select technical advisory work on systems that need to survive real business operations.

Start a conversation