Skip to content
Book a Call

AI Systems at Work

Five examples of AI systems I designed and built, including the workflows, review steps and outcomes. Client details are anonymized.

B2B Marketing Agency

An AI System for a Whole Agency Team

Tools

  • Claude
  • Claude Code
  • Claude plugins
  • Confluence
  • Jira
  • Basecamp
  • Zapier

How the Work Runs

    • Skill

      How to do the work

    • Client Context

      Who the work is for

  2. AI Writes the Draft

    Same skill, any client

  3. A Person Reviews and Decides

    Nothing ships without it

  4. Client-Ready Work

How a Skill Goes Live

  1. Request Form

  2. 1. Draft

  3. 2. Self-Test

  4. 3. Independent Test

    Not by the builder

  5. 4. Approval

  6. Library and Registry

    Owner and status

Start

A B2B marketing agency wanted every department using AI on real client work. Writing first drafts was the bottleneck. There was no shared way to build AI tools, check their quality or decide who approves them.

What I Built

  • A shared library of AI skills for recurring work: copywriting, SEO audits, campaign builds, reporting and project scoping.
  • Context files for each client, so every skill knows who the work is for.
  • A four-gate review for every skill: draft, self-test, a test by someone who didn’t build it and final approval.
  • A registry in the project tracker with each skill’s owner and status.
  • A request form, start-here guides and hands-on training for each department.

Results

  • Usage typically grows from one early adopter to at least 20 people within three to six months.
  • Each department works from a shared set of tested skills.
  • No skill reaches the team until someone other than its builder has tested it.
  • Every skill has an owner and a status in one place.
  • AI writes the first drafts. People review them and make the calls.

Marko helped us build AI services we could offer under our own brand, and taught us along the way. He knows both the AI and the engineering behind it, but makes everything easy to understand. Easy to work with, generous with his knowledge, and finished ahead of schedule.

Owner, Mid-Sized AgencyWhite-label AI services for their clients, later bought outright, plus team training. They now offer AI marketing.

Large Content Agency

A Content Pipeline for Many Clients

Tools

  • Claude
  • Claude agents
  • Claude plugins
  • Claude skills

Research and Brief

  1. Research

  2. Keyword Research

  3. AEO Research

  4. Brief

    Built from the research

Draft and Review

  1. Draft

    Written from the brief

  2. Staging

    Ready for review

  3. One Reviewer

    Approves every piece

Start

A large agency produced content for many different clients. Every piece needed research, keyword research, AEO research, a brief and a draft before anyone could review it, and the work took a team of six.

What I Built

  • Claude agents, plugins and skills for each stage of the work.
  • Research, keyword research and AEO research for every piece.
  • Briefs built from that research.
  • Drafts written from the brief.
  • Content staged and ready for review.

Results

  • Content that took a team of six now needs one reviewer.
  • About $180,000 a year in payroll and contractor costs removed, based on about $3,000 a month for each of five roles.
  • Every piece still goes through a person before it ships.

B2B Sales Team

From Booked Meeting to Second Meeting

Tools

  • Email
  • Zapier
  • HubSpot
  • Google Drive
  • Slack

Before the First Meeting

  1. Meeting Booked

  2. Company Research

    AI tools and MCP connectors

  3. Folders and Documents

    From templates in Google Drive

  4. Right People Notified

    In the apps they use

  5. Ready for the First Meeting

After the First Meeting

  1. Call Recording

  2. Custom Proof Points

    Built from the call

  3. Follow-Up Email

    Within 30 minutes of the call

  4. 95% Book a Second Meeting

Start

Every booked sales meeting needed prep: research on the company, a set of working documents and a heads-up to the right people. After the call, the follow-up had to reach the prospect while the conversation was still fresh.

What I Built

  • When a meeting is booked, AI researches the company across several AI tools and MCP connectors.
  • It creates the prospect’s folders and documents in Google Drive from templates.
  • It notifies the right people in the systems and apps they already use.
  • After the first meeting, it uses the call recording to build custom proof points and a follow-up email.

Results

  • The team walks into every first meeting prepared.
  • The follow-up goes out within 30 minutes of the call.
  • 95% of prospects book a second meeting, up from 20% in the six months before, with about the same number of leads.

Mid-Size Business

One Connected System for a Growing Business

Tools

  • HubSpot
  • Zapier
  • Claude

Request to Quote

  1. Request Comes In

  2. HubSpot

    Every request in one place

  3. Zapier

    Automations between tools

  4. Claude

    Writing, summaries and review

  5. Quote Sent

    Run by one person

Paperwork

  1. Paper Forms

  2. Digitized

    Forms and records

  3. One System

    Every record in one place

Start

A mid-size business ran its internal processes by hand, from requests for quotes to paperwork that still lived on paper. Work moved between people and tools one step at a time.

What I Built

  • HubSpot as the one place for customers, requests and quotes.
  • A request-to-quote process that runs from the first inquiry to the quote without manual handoffs.
  • Zapier automations that connect HubSpot with the rest of their tools.
  • Claude built into the steps that need writing, summaries or review.
  • Paperwork digitized, so forms and records live in one system.

Results

  • One person now handles work that used to take ten.
  • A separate technical team saves at least 10 hours a week from one process optimization.
  • Requests, quotes and records live in one place.

Analytics, Search and Indexing

Custom Connectors for Google Cloud APIs

Tools

  • Python
  • Google Cloud Run
  • MCP
  • Google Analytics 4
  • Search Console
  • Indexing API

Each Request

  1. AI Assistant

    Asks a question or runs a task

  2. Custom MCP Connectors

    Python on Google Cloud Run

  3. Service Account

    Google sign-in

    • Google Analytics 4

    • Search Console

    • Indexing API

    • Other Google APIs

Start

The team’s AI assistant couldn’t reach Google’s analytics, search and indexing tools. Every question or task started with working in those tools by hand.

What I Built

  • Custom MCP connectors for Google Cloud APIs, including Google Analytics 4, Google Search Console and the Indexing API.
  • Written in Python and hosted on Google Cloud Run as remote servers.
  • Google service account sign-in.
  • Property and site are inputs on each request, so one setup serves many sites.

Results

  • The assistant works with first-party analytics and search data directly.
  • Analytics, search and indexing tasks run from one assistant.
  • One setup covers every property the service account can access.

Want a System Like These?

Tell me which process takes your team the most time. I will show you what we could build for it.

Book a CallSend Inquiry