AI Deployment Playbook · Felipe Engineer-Manriquez · EBFC AI
EBFC.AI
ebfc.ai
The Core Insight
Build a workspace, not a chat habit.
The problem isn't the model — it's the method. Every time you open a new chat window you start over: re-explain your role, re-paste your project context, edit out the robot-speak. A workspace folder fixes that permanently. You build it once. The AI reads it every session. The folder is the system.
Chatbot vs. Workspace — The Difference
❌ Chatbot Loop
Open new chat → explain your role again → paste your project context again → get generic answer → edit out the robot-speak → close tab.
Repeat every single session. Forever.
✓ Workspace
Open workspace → AI already knows your role, your project, your tone, and what good output looks like → first draft sounds like you wrote it.
Same model. Better method. Built once, runs forever.
📁 SKILL.md — your role & rules📁 references/ — what to avoid📁 examples/ — what good looks like
Your 4-Step Deployment Plan
1
Day 1
Pick One Repetitive Task
Start narrow. Find the task that eats 30–60 min/day and produces forgettable output — daily reports, RFI responses, coordination emails. That's your proving ground.
Must be text-based and currently done by hand
Daily reports, RFI drafts, meeting recaps
Subcontractor coordination emails
Safety observation write-ups
2
Day 2–3
Build Your Workspace Folder
Create a folder — local or cloud — with three files. This is the memory you give AI. It turns a generic chatbot into your construction companion without re-explaining anything.
SKILL.md — your role, project, team, communication tone
references/ — phrases to avoid, patterns that sound robotic
examples/ — one good RFI, one clean report, one email that landed
Point your AI tool at this folder once. Done.
3
Day 4–5
Build Your Prompt Process as a .md File
Document the exact steps you follow to get a good output for your chosen task. Save it as a Markdown file in your workspace. Now anyone on your team can run it in 60 seconds.
Format: Role → Context → Raw Input → Output Format
Name it by task: daily-report.md, rfi-draft.md
Drop it in your workspace folder — AI reads it automatically
Refine it when outputs need adjustment — the file is the system
4
Week 2 → Ongoing
Run 2 Weeks, Edit Outputs, Refine
AI output is a first draft, not a final one. Edit like a superintendent reviews a punchlist — inspect the work, then fix the .md file so the same mistake doesn't come back.
Run your prompt process every day for 2 weeks
Track your edits — patterns in your fixes = prompt improvements
Update SKILL.md with anything you keep having to correct
Share the folder with your team — they inherit your system, not your effort
The Workspace Formula — Quick Reference
1
Start a folder
One place: role, project, rules
2
Write your rules
Save as SKILL.md in the folder
3
Add examples
One good RFI, report, email
4
Let it work
AI reads the folder every session
From EBFC AI · For You
Two files. Everything you need to start Monday.
No email required. No account. Open either file in any app — or paste the prompt directly into Claude or ChatGPT right now.
📋
The Playbook
The full 4-step workspace system as a Markdown file. Works in Notion, Obsidian, VS Code, Apple Notes — any app that reads text. Your reference as you build.
Open this file, copy everything, paste into Claude or ChatGPT. Your AI writes like a field professional from the first message. No setup. Works right now.