Code3 is a digital advertising agency. I ran its coaching sprint on an “I do, we do, you do” model, and within a quarter, in the client’s words, teams across the business were building and maintaining their own AI-powered tools.
20hours
70%
of hands-on coaching, sponsored by the CFO
average adoption within each department, measured by weekly activity with the Claude Skills, Slack bots and automations
“Rasheed worked closely with my team to determine where AI deployment could help us both short-term and long-term. He helped migrate our organization to Claude, build specialized Skill packages, and integrate those skills with Slack and Asana. Within a quarter we had teams across our business building and maintaining their own AI-powered tools. Rasheed helped put us on a path to continue building independently, and months later we are still developing new ways to implement AI.”
Code3 is a 140-person digital advertising agency. Its teams worked in ChatGPT, with Custom GPTs of their own, and the work ahead included moving the organization to Claude.
Its CFO sponsored a 20-hour coaching sprint. Jack Gilbert, Director of Performance & Analytics, gives the brief in his own words: to “determine where AI deployment could help us both short-term and long-term.”
The sprint was contracted as coaching, and every piece of it ran on one model. I do it while the team watches. We do it together. They do it, and keep doing it, without me. The point of the work was the hand-off.
Grouped by who held the tool, not by date.
The groundwork I took on myself, so the team had something solid to pick up.
The tools the teams use every day, built with them rather than for them.
Teams across the agency building and maintaining their own AI-powered tools, under rules they can run without me.
The move off ChatGPT: a Drive knowledge base, Cowork as the daily workspace, a Custom GPT-to-Skills migration path, the tradeoffs stated, and a 30/60/90-day rollout.
Lets non-engineers build and govern Claude Skills, with reusable templates and ownership, versioning, quarterly-audit and sunset rules.
Used by non-engineers.
For every department’s workflows, connected to Asana, Google Workspace and Figma.
Used by marketing, client relations, data and performance, and creative.
Retrieval over the agency’s documents on Postgres with pgvector, with agentic retrieval and LLM reranking, access controls, and continuous ingestion of new documents and messages.
Scored on ROI, build effort, and adoption and vendor risk, for executive review. It re-ranked the agency’s priorities, added two themes and laid out a 30/60/90-day plan.
Used by executives, for review.
Contents
Migration playbook
The rules
Skill-creation playbook
So non-engineers can build and govern Claude Skills.
Two company-wide AI office hours, about 20 attendees each, turned into a five-theme gap analysis.
An average of 70% adoption within each department, measured by weekly activity with the Claude Skills, Slack bots and automations.
“Rasheed helped put us on a path to continue building independently, and months later we are still developing new ways to implement AI.”
Built with the teams, kept by the teams.
Claude/ Claude Skills/ Slack/ Asana/ Google Workspace/ Figma/ MCP/ Postgres/ pgvector/ Cowork/ Google Drive
