AI Immersion — Jump 450

The Idea

AI Immersion is a one-day onboarding that takes every new hire from zero to shipping a working app with Claude Code. By end of day they have a fully provisioned AI toolkit, a demoed solution to a real Jump 450 pain point, and a rubric score that calibrates their follow-up support.

Everyone goes through it — buyers, creative, PMs, MDs. Challenges are assigned by role, so the bar stays the same while the problem stays relevant.

Running the day? This page is the organizer playbook. Send your hire the new-hire guide — same day, written for them.

Panel: log rubric scores in the Immersion Scorecard — shared with the panel only, hires never see it.

Roles

AI Buddy

Teammate from the hire's discipline. Runs morning provisioning, assigns the challenge, does the midday check-in, first line of help.

Support Backstop

#internal_ai-support admins. Unblock provisioning failures the AI Buddy can't fix.

Review Panel

AI Buddy + hire's manager (+1 optional). Attends the demo, scores the rubric, delivers feedback.

Before day one · prep

Before the Day

Nothing here happens live — get all four done before the hire arrives so the morning is spent building, not scrambling.

Hire has laptop, Slack, and email active (Glass signs in via Slack — workspace jump-450).
AI Buddy is assigned and calendar holds are sent.
A role-matched challenge is picked.
The hire's Claude seat is confirmed provisioned.

The day itself · 10:00 – 4:00

Run of Day

10:00 – 10:45

Kickoff + provisioning Live with AI Buddy

Sign in to glass.jump450.com and work through setup: Claude Code one-command install (Xcode/Node, GitHub auth, Doppler, Vercel CLI) and Claude Desktop. Verify with the smoke test.

10:45 – 11:00

Challenge briefing

AI Buddy assigns the pain point, walks through expectations and the rubric, points to the Glass prompt starter pack and dojo.jump450.com.

11:00 – 12:30

Build block 1

Solo with Claude Code. Posting questions in #internal_ai-support is encouraged — asking well is part of the skill.

12:30 – 1:00

Midday check-in AI Buddy

Review progress, unblock, rescope if needed.

1:00 – 1:30

Lunch / reset

A real break — the hire steps away and eats before the final build. Don't skip it.

1:30 – 3:00

Build block 2

Keep building; deploy if possible — Vercel is already configured from setup.

3:00 – 3:15

Demo prep

Five-minute walkthrough: the problem, what they built, how they worked with Claude.

3:15 – 3:45

Demo + review Panel

5-minute demo, 10-minute Q&A, panel scores the rubric live.

3:45 – 4:00

Debrief

Scores and feedback; agree next steps — Dojo skills to adopt, a follow-up build session if needed.

The rest of this guide zooms into the four moments that need detail — the morning gate, the challenge, the check-in, and the review — in the order they happen.

10:00 – 10:45 · Kickoff gate

Morning Smoke Test

Provisioning isn't done until every box checks. The AI Buddy verifies each item live:

Glass — signed in via Slack, setup checklist complete
Claude Desktop — opens, signed in to the Jump 450 org, sends a message
Claude Code — runs in terminal; hire prompts it to create and run a hello-world script
GitHub — authenticated; hire can clone a repo
Doppler + Vercel CLI — installed and authenticated
Slack — hire has joined #internal_ai-support

10:45 – 11:00 · Briefing

Challenge Prompt Bank

One per hire, matched to role. Good challenges are real pain points, scoped to a day, and demoable.

Role	Sample challenge
Project Manager	An intake form + tracker app that turns ad-hoc Slack requests into a prioritized queue with statuses and owners.
Creative Director	An ad-copy variant generator that takes a brief and brand voice notes and outputs testable variants in a review UI.
Motion Graphics Director	A deliverables spec generator that takes one master concept and outputs platform-ready cutdown lists (aspect ratios, durations, safe zones) for Meta, TikTok, and YouTube.
Google Buyer	A search-terms audit tool that ingests a search terms report CSV, flags wasted spend, and suggests negative keywords.
Social Buyer	A creative fatigue tracker that ingests a Meta ads export and flags ads with rising frequency and declining CTR that need a refresh.
Programmatic Buyer	A placement audit tool that ingests a placement report and builds an exclusion list of low-viewability or junk domains.
Managing Director	A client health one-pager that combines pacing and performance from account exports into a weekly briefing view.

Scope guidance: a working slice beats a broken vision. Steer hires toward something demoable by 3:00 — treat cut scope as judgment, not failure.

12:30 – 1:00 · Check-in

Midday Check-in Script

Show me what you have so far — even if it's broken.
What has Claude Code been great at? Where has it fought you?
Show me your last few prompts — let's look at how you're asking.
What's the demo at 3:15 if you stopped right now? What's the one thing to add next?
Anything blocked that I or #internal_ai-support should fix right now?

3:15 – 3:45 · Review

Scoring Rubric

Each panelist scores 1–4 per criterion using the anchors below; scores are averaged across panelists, then weighted. Evidence comes from the demo, Q&A, and the AI Buddy's midday observation of the hire's prompts.

Working output 30%

Nothing runs; fragments or a plan only.
Partial: core pieces work but the demo path breaks.
Runs end-to-end on the assigned pain point, rough edges.
Runs end-to-end; deployed or trivially runnable by someone else.

AI fluency 30%

Vague prompts; accepts first output; no iteration.
Some specificity; iterates, but by trial-and-error.
Specific, context-rich prompts; course-corrects Claude deliberately.
All of 3, plus decomposes the problem into steps and leverages Glass starter prompts or Dojo skills.

Scope & judgment 20%

Chased the full vision; nothing demoable; never asked for help.
Rescoped only when pushed at the check-in; asked for help late.
Prioritized a demoable slice; asked for help when stuck.
Cut the right corners early and unprompted; sequenced work demo-first; escalated blockers fast.

Communication 20%

Demo unclear; can't explain what was built or how.
Shows the app but the problem or the AI process is muddled.
Clear problem → solution → AI-process story; honest about limitations.
All of 3, crisp within time, and handles Q&A with specific prompts and decisions from their session.

Score Tiers

3.4 – 4.0

AI-native

Fast-track — invite to contribute a skill to Dojo.

2.5 – 3.3

On track

Standard follow-up — one 1:1 build session within 30 days.

1.0 – 2.4

Needs support

Two 1:1 build sessions within 30 days.

What's a 1:1 build session? A scheduled hour where the hire and their AI Buddy work a real task from the hire's actual job together in Claude Code. The hire drives; the buddy coaches prompting and workflow in the moment. It ends with one concrete habit to practice before the next session.

Scores calibrate support, not standing. Everyone is scored on the same rubric — the challenge, not the bar, is what's role-adjusted.

Troubleshooting & Escalation

First stop: the troubleshooting list on glass.jump450.com — a running list of known fixes from pilots.
Second: #internal_ai-support — the AI Buddy posts on the hire's behalf if they're stuck more than 20 minutes.
Provisioning failure that can't be fixed by 11:00: reschedule the hands-on build rather than losing the day to setup — the hire can still do a lighter, chat-only version of the challenge in the meantime.

After the Day

AI Buddy logs scores and one paragraph of notes; shares with the hire's manager.
Hire is pointed to dojo.jump450.com for shared skills and websites to build on.
Demo artifacts are collected — the best ones become future challenge examples.
Follow-up cadence per score tier.