NY Tech Week 2026 · Thursday June 4, 2026 · Chelsea, New York
Build an Agentic Team with Claude Code
An evening workshop co-led by Nicole Patten and Harry Roper. By 7pm the room had an agentic team running on their own laptops: a watcher that scrapes the web for what matters in their world, scores it, and drafts content in their voice. They built it by describing what they wanted to Claude Code. No coding required.





The build
What the room built
The room ran seven prompts in order, and each one added an agent to the same repo. The room moved on when 80% of it was ready, and helpers on the floor unstuck anyone behind for more than three minutes.
01
The structure prompt
Pasted into Claude Code first. It sets up a two-folder framework, directives that describe and executions that run, and a CLAUDE.md that orchestrates them.
02
The Scout
The kickoff prompt. Each attendee named their own sources and topic, as specific as they could make it, and had Claude Code plan the watcher before writing any code.
03
Claude writes its own briefing
With the Scout running, one slash command inside Claude Code read everything on disk and wrote the CLAUDE.md that orchestrated the rest of the night.
04
The Scorer
Reads what the Scout found and ranks it, built with the same prompt shape as the Scout.
05
The Editor
Takes the top-scored items and turns them into briefs. Every prompt from here on ended the same way: keep the orchestration layer current.
06
The Writer
Drafts a post from each brief, in the attendee's own voice, from the two or three paragraphs of their writing they brought with them.
07
The dashboard
Claude Code pointed at the final output and asked to build a dashboard in the same editor.
Try it
The first two prompts
These are the prompts the room started from, as written. Open a new folder in Claude Code, paste the first, then replace the bracketed parts of the second with your own sources and topic.
01 · The structure prompt
Set up an agentic workflow framework with two folders: - /directives - markdown files describing what each workflow does and which scripts to run - /executions - standalone Python scripts that do the actual work Add a CLAUDE.md that tells the agent: read the directive, run the scripts, report results. Use .env for API keys, python-dotenv to load them.
02 · The Scout
Hello, build me a watcher that: - monitors [my sources] for [my topic] - pulls the last 24 hours - returns top 10: title, link, summary, date - saves results to a local file CLAUDE.md should act as the orchestration layer - it reads first and routes every directive to the right execution script. First, search the web for scraping tools. I have a Firecrawl key in .env. Give me 3 options + a plan. No code yet.
By 7pm
What attendees walked out with
- A working watcher running locally on their laptop
- A structured repo: directives, orchestration, execution
- A dashboard built in the same editor
- A drafted piece of content they could publish that night
- The mental model to add workflow #2 through #20 to the same repo
The team
Who led it


Run this session for your room
It runs as an evening workshop or a half day, for a company team or a conference room, with no coding experience required. Send the event, the date and the audience and Nicole replies with availability and a quote.